1 // SPDX-License-Identifier: GPL-2.0-only <<
2 /* 1 /*
3 * linux/fs/exec.c 2 * linux/fs/exec.c
4 * 3 *
5 * Copyright (C) 1991, 1992 Linus Torvalds 4 * Copyright (C) 1991, 1992 Linus Torvalds
6 */ 5 */
7 6
8 /* 7 /*
9 * #!-checking implemented by tytso. 8 * #!-checking implemented by tytso.
10 */ 9 */
11 /* 10 /*
12 * Demand-loading implemented 01.12.91 - no ne 11 * Demand-loading implemented 01.12.91 - no need to read anything but
13 * the header into memory. The inode of the ex 12 * the header into memory. The inode of the executable is put into
14 * "current->executable", and page faults do t 13 * "current->executable", and page faults do the actual loading. Clean.
15 * 14 *
16 * Once more I can proudly say that linux stoo 15 * Once more I can proudly say that linux stood up to being changed: it
17 * was less than 2 hours work to get demand-lo 16 * was less than 2 hours work to get demand-loading completely implemented.
18 * 17 *
19 * Demand loading changed July 1993 by Eric Yo 18 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
20 * current->executable is only used by the pro 19 * current->executable is only used by the procfs. This allows a dispatch
21 * table to check for several different types 20 * table to check for several different types of binary formats. We keep
22 * trying until we recognize the file or we ru 21 * trying until we recognize the file or we run out of supported binary
23 * formats. 22 * formats.
24 */ 23 */
25 24
26 #include <linux/kernel_read_file.h> <<
27 #include <linux/slab.h> 25 #include <linux/slab.h>
28 #include <linux/file.h> 26 #include <linux/file.h>
29 #include <linux/fdtable.h> 27 #include <linux/fdtable.h>
30 #include <linux/mm.h> 28 #include <linux/mm.h>
>> 29 #include <linux/vmacache.h>
31 #include <linux/stat.h> 30 #include <linux/stat.h>
32 #include <linux/fcntl.h> 31 #include <linux/fcntl.h>
33 #include <linux/swap.h> 32 #include <linux/swap.h>
34 #include <linux/string.h> 33 #include <linux/string.h>
35 #include <linux/init.h> 34 #include <linux/init.h>
36 #include <linux/sched/mm.h> <<
37 #include <linux/sched/coredump.h> <<
38 #include <linux/sched/signal.h> <<
39 #include <linux/sched/numa_balancing.h> <<
40 #include <linux/sched/task.h> <<
41 #include <linux/pagemap.h> 35 #include <linux/pagemap.h>
42 #include <linux/perf_event.h> 36 #include <linux/perf_event.h>
43 #include <linux/highmem.h> 37 #include <linux/highmem.h>
44 #include <linux/spinlock.h> 38 #include <linux/spinlock.h>
45 #include <linux/key.h> 39 #include <linux/key.h>
46 #include <linux/personality.h> 40 #include <linux/personality.h>
47 #include <linux/binfmts.h> 41 #include <linux/binfmts.h>
48 #include <linux/utsname.h> 42 #include <linux/utsname.h>
49 #include <linux/pid_namespace.h> 43 #include <linux/pid_namespace.h>
50 #include <linux/module.h> 44 #include <linux/module.h>
51 #include <linux/namei.h> 45 #include <linux/namei.h>
52 #include <linux/mount.h> 46 #include <linux/mount.h>
53 #include <linux/security.h> 47 #include <linux/security.h>
54 #include <linux/syscalls.h> 48 #include <linux/syscalls.h>
55 #include <linux/tsacct_kern.h> 49 #include <linux/tsacct_kern.h>
56 #include <linux/cn_proc.h> 50 #include <linux/cn_proc.h>
57 #include <linux/audit.h> 51 #include <linux/audit.h>
>> 52 #include <linux/tracehook.h>
58 #include <linux/kmod.h> 53 #include <linux/kmod.h>
59 #include <linux/fsnotify.h> 54 #include <linux/fsnotify.h>
60 #include <linux/fs_struct.h> 55 #include <linux/fs_struct.h>
>> 56 #include <linux/pipe_fs_i.h>
61 #include <linux/oom.h> 57 #include <linux/oom.h>
62 #include <linux/compat.h> 58 #include <linux/compat.h>
63 #include <linux/vmalloc.h> 59 #include <linux/vmalloc.h>
64 #include <linux/io_uring.h> <<
65 #include <linux/syscall_user_dispatch.h> <<
66 #include <linux/coredump.h> <<
67 #include <linux/time_namespace.h> <<
68 #include <linux/user_events.h> <<
69 #include <linux/rseq.h> <<
70 #include <linux/ksm.h> <<
71 60
72 #include <linux/uaccess.h> 61 #include <linux/uaccess.h>
73 #include <asm/mmu_context.h> 62 #include <asm/mmu_context.h>
74 #include <asm/tlb.h> 63 #include <asm/tlb.h>
75 64
76 #include <trace/events/task.h> 65 #include <trace/events/task.h>
77 #include "internal.h" 66 #include "internal.h"
78 67
79 #include <trace/events/sched.h> 68 #include <trace/events/sched.h>
80 69
81 static int bprm_creds_from_file(struct linux_b <<
82 <<
83 int suid_dumpable = 0; 70 int suid_dumpable = 0;
84 71
85 static LIST_HEAD(formats); 72 static LIST_HEAD(formats);
86 static DEFINE_RWLOCK(binfmt_lock); 73 static DEFINE_RWLOCK(binfmt_lock);
87 74
88 void __register_binfmt(struct linux_binfmt * f 75 void __register_binfmt(struct linux_binfmt * fmt, int insert)
89 { 76 {
>> 77 BUG_ON(!fmt);
>> 78 if (WARN_ON(!fmt->load_binary))
>> 79 return;
90 write_lock(&binfmt_lock); 80 write_lock(&binfmt_lock);
91 insert ? list_add(&fmt->lh, &formats) 81 insert ? list_add(&fmt->lh, &formats) :
92 list_add_tail(&fmt->lh, &form 82 list_add_tail(&fmt->lh, &formats);
93 write_unlock(&binfmt_lock); 83 write_unlock(&binfmt_lock);
94 } 84 }
95 85
96 EXPORT_SYMBOL(__register_binfmt); 86 EXPORT_SYMBOL(__register_binfmt);
97 87
98 void unregister_binfmt(struct linux_binfmt * f 88 void unregister_binfmt(struct linux_binfmt * fmt)
99 { 89 {
100 write_lock(&binfmt_lock); 90 write_lock(&binfmt_lock);
101 list_del(&fmt->lh); 91 list_del(&fmt->lh);
102 write_unlock(&binfmt_lock); 92 write_unlock(&binfmt_lock);
103 } 93 }
104 94
105 EXPORT_SYMBOL(unregister_binfmt); 95 EXPORT_SYMBOL(unregister_binfmt);
106 96
107 static inline void put_binfmt(struct linux_bin 97 static inline void put_binfmt(struct linux_binfmt * fmt)
108 { 98 {
109 module_put(fmt->module); 99 module_put(fmt->module);
110 } 100 }
111 101
112 bool path_noexec(const struct path *path) 102 bool path_noexec(const struct path *path)
113 { 103 {
114 return (path->mnt->mnt_flags & MNT_NOE 104 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
115 (path->mnt->mnt_sb->s_iflags & 105 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
116 } 106 }
117 107
118 #ifdef CONFIG_USELIB 108 #ifdef CONFIG_USELIB
119 /* 109 /*
120 * Note that a shared library must be both rea 110 * Note that a shared library must be both readable and executable due to
121 * security reasons. 111 * security reasons.
122 * 112 *
123 * Also note that we take the address to load !! 113 * Also note that we take the address to load from from the file itself.
124 */ 114 */
125 SYSCALL_DEFINE1(uselib, const char __user *, l 115 SYSCALL_DEFINE1(uselib, const char __user *, library)
126 { 116 {
127 struct linux_binfmt *fmt; 117 struct linux_binfmt *fmt;
128 struct file *file; 118 struct file *file;
129 struct filename *tmp = getname(library 119 struct filename *tmp = getname(library);
130 int error = PTR_ERR(tmp); 120 int error = PTR_ERR(tmp);
131 static const struct open_flags uselib_ 121 static const struct open_flags uselib_flags = {
132 .open_flag = O_LARGEFILE | O_R !! 122 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
133 .acc_mode = MAY_READ | MAY_EXE 123 .acc_mode = MAY_READ | MAY_EXEC,
134 .intent = LOOKUP_OPEN, 124 .intent = LOOKUP_OPEN,
135 .lookup_flags = LOOKUP_FOLLOW, 125 .lookup_flags = LOOKUP_FOLLOW,
136 }; 126 };
137 127
138 if (IS_ERR(tmp)) 128 if (IS_ERR(tmp))
139 goto out; 129 goto out;
140 130
141 file = do_filp_open(AT_FDCWD, tmp, &us 131 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
142 putname(tmp); 132 putname(tmp);
143 error = PTR_ERR(file); 133 error = PTR_ERR(file);
144 if (IS_ERR(file)) 134 if (IS_ERR(file))
145 goto out; 135 goto out;
146 136
147 /* !! 137 error = -EINVAL;
148 * Check do_open_execat() for an expla !! 138 if (!S_ISREG(file_inode(file)->i_mode))
149 */ !! 139 goto exit;
>> 140
150 error = -EACCES; 141 error = -EACCES;
151 if (WARN_ON_ONCE(!S_ISREG(file_inode(f !! 142 if (path_noexec(&file->f_path))
152 path_noexec(&file->f_path)) <<
153 goto exit; 143 goto exit;
154 144
>> 145 fsnotify_open(file);
>> 146
155 error = -ENOEXEC; 147 error = -ENOEXEC;
156 148
157 read_lock(&binfmt_lock); 149 read_lock(&binfmt_lock);
158 list_for_each_entry(fmt, &formats, lh) 150 list_for_each_entry(fmt, &formats, lh) {
159 if (!fmt->load_shlib) 151 if (!fmt->load_shlib)
160 continue; 152 continue;
161 if (!try_module_get(fmt->modul 153 if (!try_module_get(fmt->module))
162 continue; 154 continue;
163 read_unlock(&binfmt_lock); 155 read_unlock(&binfmt_lock);
164 error = fmt->load_shlib(file); 156 error = fmt->load_shlib(file);
165 read_lock(&binfmt_lock); 157 read_lock(&binfmt_lock);
166 put_binfmt(fmt); 158 put_binfmt(fmt);
167 if (error != -ENOEXEC) 159 if (error != -ENOEXEC)
168 break; 160 break;
169 } 161 }
170 read_unlock(&binfmt_lock); 162 read_unlock(&binfmt_lock);
171 exit: 163 exit:
172 fput(file); 164 fput(file);
173 out: 165 out:
174 return error; !! 166 return error;
175 } 167 }
176 #endif /* #ifdef CONFIG_USELIB */ 168 #endif /* #ifdef CONFIG_USELIB */
177 169
178 #ifdef CONFIG_MMU 170 #ifdef CONFIG_MMU
179 /* 171 /*
180 * The nascent bprm->mm is not visible until e 172 * The nascent bprm->mm is not visible until exec_mmap() but it can
181 * use a lot of memory, account these pages in 173 * use a lot of memory, account these pages in current->mm temporary
182 * for oom_badness()->get_mm_rss(). Once exec 174 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
183 * change the counter back via acct_arg_size(0 175 * change the counter back via acct_arg_size(0).
184 */ 176 */
185 static void acct_arg_size(struct linux_binprm 177 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
186 { 178 {
187 struct mm_struct *mm = current->mm; 179 struct mm_struct *mm = current->mm;
188 long diff = (long)(pages - bprm->vma_p 180 long diff = (long)(pages - bprm->vma_pages);
189 181
190 if (!mm || !diff) 182 if (!mm || !diff)
191 return; 183 return;
192 184
193 bprm->vma_pages = pages; 185 bprm->vma_pages = pages;
194 add_mm_counter(mm, MM_ANONPAGES, diff) 186 add_mm_counter(mm, MM_ANONPAGES, diff);
195 } 187 }
196 188
197 static struct page *get_arg_page(struct linux_ 189 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
198 int write) 190 int write)
199 { 191 {
200 struct page *page; 192 struct page *page;
201 struct vm_area_struct *vma = bprm->vma <<
202 struct mm_struct *mm = bprm->mm; <<
203 int ret; 193 int ret;
>> 194 unsigned int gup_flags = FOLL_FORCE;
204 195
205 /* !! 196 #ifdef CONFIG_STACK_GROWSUP
206 * Avoid relying on expanding the stac !! 197 if (write) {
207 * does not work for STACK_GROWSUP any !! 198 ret = expand_downwards(bprm->vma, pos);
208 * by hand ahead of time. !! 199 if (ret < 0)
209 */ <<
210 if (write && pos < vma->vm_start) { <<
211 mmap_write_lock(mm); <<
212 ret = expand_downwards(vma, po <<
213 if (unlikely(ret < 0)) { <<
214 mmap_write_unlock(mm); <<
215 return NULL; 200 return NULL;
216 } !! 201 }
217 mmap_write_downgrade(mm); !! 202 #endif
218 } else !! 203
219 mmap_read_lock(mm); !! 204 if (write)
>> 205 gup_flags |= FOLL_WRITE;
220 206
221 /* 207 /*
222 * We are doing an exec(). 'current' 208 * We are doing an exec(). 'current' is the process
223 * doing the exec and 'mm' is the new !! 209 * doing the exec and bprm->mm is the new process's mm.
224 */ 210 */
225 ret = get_user_pages_remote(mm, pos, 1 !! 211 ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
226 write ? FOLL_WRITE : 0 !! 212 &page, NULL, NULL);
227 &page, NULL); <<
228 mmap_read_unlock(mm); <<
229 if (ret <= 0) 213 if (ret <= 0)
230 return NULL; 214 return NULL;
231 215
232 if (write) !! 216 if (write) {
233 acct_arg_size(bprm, vma_pages( !! 217 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
>> 218 struct rlimit *rlim;
>> 219
>> 220 acct_arg_size(bprm, size / PAGE_SIZE);
>> 221
>> 222 /*
>> 223 * We've historically supported up to 32 pages (ARG_MAX)
>> 224 * of argument strings even with small stacks
>> 225 */
>> 226 if (size <= ARG_MAX)
>> 227 return page;
>> 228
>> 229 /*
>> 230 * Limit to 1/4-th the stack size for the argv+env strings.
>> 231 * This ensures that:
>> 232 * - the remaining binfmt code will not run out of stack space,
>> 233 * - the program will have a reasonable amount of stack left
>> 234 * to work from.
>> 235 */
>> 236 rlim = current->signal->rlim;
>> 237 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
>> 238 put_page(page);
>> 239 return NULL;
>> 240 }
>> 241 }
234 242
235 return page; 243 return page;
236 } 244 }
237 245
238 static void put_arg_page(struct page *page) 246 static void put_arg_page(struct page *page)
239 { 247 {
240 put_page(page); 248 put_page(page);
241 } 249 }
242 250
243 static void free_arg_pages(struct linux_binprm 251 static void free_arg_pages(struct linux_binprm *bprm)
244 { 252 {
245 } 253 }
246 254
247 static void flush_arg_page(struct linux_binprm 255 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
248 struct page *page) 256 struct page *page)
249 { 257 {
250 flush_cache_page(bprm->vma, pos, page_ 258 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
251 } 259 }
252 260
253 static int __bprm_mm_init(struct linux_binprm 261 static int __bprm_mm_init(struct linux_binprm *bprm)
254 { 262 {
255 int err; 263 int err;
256 struct vm_area_struct *vma = NULL; 264 struct vm_area_struct *vma = NULL;
257 struct mm_struct *mm = bprm->mm; 265 struct mm_struct *mm = bprm->mm;
258 266
259 bprm->vma = vma = vm_area_alloc(mm); !! 267 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
260 if (!vma) 268 if (!vma)
261 return -ENOMEM; 269 return -ENOMEM;
262 vma_set_anonymous(vma); <<
263 270
264 if (mmap_write_lock_killable(mm)) { !! 271 if (down_write_killable(&mm->mmap_sem)) {
265 err = -EINTR; 272 err = -EINTR;
266 goto err_free; 273 goto err_free;
267 } 274 }
268 !! 275 vma->vm_mm = mm;
269 /* <<
270 * Need to be called with mmap write l <<
271 * held, to avoid race with ksmd. <<
272 */ <<
273 err = ksm_execve(mm); <<
274 if (err) <<
275 goto err_ksm; <<
276 276
277 /* 277 /*
278 * Place the stack at the largest stac 278 * Place the stack at the largest stack address the architecture
279 * supports. Later, we'll move this to 279 * supports. Later, we'll move this to an appropriate place. We don't
280 * use STACK_TOP because that can depe 280 * use STACK_TOP because that can depend on attributes which aren't
281 * configured yet. 281 * configured yet.
282 */ 282 */
283 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK 283 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
284 vma->vm_end = STACK_TOP_MAX; 284 vma->vm_end = STACK_TOP_MAX;
285 vma->vm_start = vma->vm_end - PAGE_SIZ 285 vma->vm_start = vma->vm_end - PAGE_SIZE;
286 vm_flags_init(vma, VM_SOFTDIRTY | VM_S !! 286 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
287 vma->vm_page_prot = vm_get_page_prot(v 287 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
>> 288 INIT_LIST_HEAD(&vma->anon_vma_chain);
288 289
289 err = insert_vm_struct(mm, vma); 290 err = insert_vm_struct(mm, vma);
290 if (err) 291 if (err)
291 goto err; 292 goto err;
292 293
293 mm->stack_vm = mm->total_vm = 1; 294 mm->stack_vm = mm->total_vm = 1;
294 mmap_write_unlock(mm); !! 295 arch_bprm_mm_init(mm, vma);
>> 296 up_write(&mm->mmap_sem);
295 bprm->p = vma->vm_end - sizeof(void *) 297 bprm->p = vma->vm_end - sizeof(void *);
296 return 0; 298 return 0;
297 err: 299 err:
298 ksm_exit(mm); !! 300 up_write(&mm->mmap_sem);
299 err_ksm: <<
300 mmap_write_unlock(mm); <<
301 err_free: 301 err_free:
302 bprm->vma = NULL; 302 bprm->vma = NULL;
303 vm_area_free(vma); !! 303 kmem_cache_free(vm_area_cachep, vma);
304 return err; 304 return err;
305 } 305 }
306 306
307 static bool valid_arg_len(struct linux_binprm 307 static bool valid_arg_len(struct linux_binprm *bprm, long len)
308 { 308 {
309 return len <= MAX_ARG_STRLEN; 309 return len <= MAX_ARG_STRLEN;
310 } 310 }
311 311
312 #else 312 #else
313 313
314 static inline void acct_arg_size(struct linux_ 314 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
315 { 315 {
316 } 316 }
317 317
318 static struct page *get_arg_page(struct linux_ 318 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
319 int write) 319 int write)
320 { 320 {
321 struct page *page; 321 struct page *page;
322 322
323 page = bprm->page[pos / PAGE_SIZE]; 323 page = bprm->page[pos / PAGE_SIZE];
324 if (!page && write) { 324 if (!page && write) {
325 page = alloc_page(GFP_HIGHUSER 325 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
326 if (!page) 326 if (!page)
327 return NULL; 327 return NULL;
328 bprm->page[pos / PAGE_SIZE] = 328 bprm->page[pos / PAGE_SIZE] = page;
329 } 329 }
330 330
331 return page; 331 return page;
332 } 332 }
333 333
334 static void put_arg_page(struct page *page) 334 static void put_arg_page(struct page *page)
335 { 335 {
336 } 336 }
337 337
338 static void free_arg_page(struct linux_binprm 338 static void free_arg_page(struct linux_binprm *bprm, int i)
339 { 339 {
340 if (bprm->page[i]) { 340 if (bprm->page[i]) {
341 __free_page(bprm->page[i]); 341 __free_page(bprm->page[i]);
342 bprm->page[i] = NULL; 342 bprm->page[i] = NULL;
343 } 343 }
344 } 344 }
345 345
346 static void free_arg_pages(struct linux_binprm 346 static void free_arg_pages(struct linux_binprm *bprm)
347 { 347 {
348 int i; 348 int i;
349 349
350 for (i = 0; i < MAX_ARG_PAGES; i++) 350 for (i = 0; i < MAX_ARG_PAGES; i++)
351 free_arg_page(bprm, i); 351 free_arg_page(bprm, i);
352 } 352 }
353 353
354 static void flush_arg_page(struct linux_binprm 354 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
355 struct page *page) 355 struct page *page)
356 { 356 {
357 } 357 }
358 358
359 static int __bprm_mm_init(struct linux_binprm 359 static int __bprm_mm_init(struct linux_binprm *bprm)
360 { 360 {
361 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - 361 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
362 return 0; 362 return 0;
363 } 363 }
364 364
365 static bool valid_arg_len(struct linux_binprm 365 static bool valid_arg_len(struct linux_binprm *bprm, long len)
366 { 366 {
367 return len <= bprm->p; 367 return len <= bprm->p;
368 } 368 }
369 369
370 #endif /* CONFIG_MMU */ 370 #endif /* CONFIG_MMU */
371 371
372 /* 372 /*
373 * Create a new mm_struct and populate it with 373 * Create a new mm_struct and populate it with a temporary stack
374 * vm_area_struct. We don't have enough conte 374 * vm_area_struct. We don't have enough context at this point to set the stack
375 * flags, permissions, and offset, so we use t 375 * flags, permissions, and offset, so we use temporary values. We'll update
376 * them later in setup_arg_pages(). 376 * them later in setup_arg_pages().
377 */ 377 */
378 static int bprm_mm_init(struct linux_binprm *b 378 static int bprm_mm_init(struct linux_binprm *bprm)
379 { 379 {
380 int err; 380 int err;
381 struct mm_struct *mm = NULL; 381 struct mm_struct *mm = NULL;
382 382
383 bprm->mm = mm = mm_alloc(); 383 bprm->mm = mm = mm_alloc();
384 err = -ENOMEM; 384 err = -ENOMEM;
385 if (!mm) 385 if (!mm)
386 goto err; 386 goto err;
387 387
388 /* Save current stack limit for all ca <<
389 task_lock(current->group_leader); <<
390 bprm->rlim_stack = current->signal->rl <<
391 task_unlock(current->group_leader); <<
392 <<
393 err = __bprm_mm_init(bprm); 388 err = __bprm_mm_init(bprm);
394 if (err) 389 if (err)
395 goto err; 390 goto err;
396 391
397 return 0; 392 return 0;
398 393
399 err: 394 err:
400 if (mm) { 395 if (mm) {
401 bprm->mm = NULL; 396 bprm->mm = NULL;
402 mmdrop(mm); 397 mmdrop(mm);
403 } 398 }
404 399
405 return err; 400 return err;
406 } 401 }
407 402
408 struct user_arg_ptr { 403 struct user_arg_ptr {
409 #ifdef CONFIG_COMPAT 404 #ifdef CONFIG_COMPAT
410 bool is_compat; 405 bool is_compat;
411 #endif 406 #endif
412 union { 407 union {
413 const char __user *const __use 408 const char __user *const __user *native;
414 #ifdef CONFIG_COMPAT 409 #ifdef CONFIG_COMPAT
415 const compat_uptr_t __user *co 410 const compat_uptr_t __user *compat;
416 #endif 411 #endif
417 } ptr; 412 } ptr;
418 }; 413 };
419 414
420 static const char __user *get_user_arg_ptr(str 415 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
421 { 416 {
422 const char __user *native; 417 const char __user *native;
423 418
424 #ifdef CONFIG_COMPAT 419 #ifdef CONFIG_COMPAT
425 if (unlikely(argv.is_compat)) { 420 if (unlikely(argv.is_compat)) {
426 compat_uptr_t compat; 421 compat_uptr_t compat;
427 422
428 if (get_user(compat, argv.ptr. 423 if (get_user(compat, argv.ptr.compat + nr))
429 return ERR_PTR(-EFAULT 424 return ERR_PTR(-EFAULT);
430 425
431 return compat_ptr(compat); 426 return compat_ptr(compat);
432 } 427 }
433 #endif 428 #endif
434 429
435 if (get_user(native, argv.ptr.native + 430 if (get_user(native, argv.ptr.native + nr))
436 return ERR_PTR(-EFAULT); 431 return ERR_PTR(-EFAULT);
437 432
438 return native; 433 return native;
439 } 434 }
440 435
441 /* 436 /*
442 * count() counts the number of strings in arr 437 * count() counts the number of strings in array ARGV.
443 */ 438 */
444 static int count(struct user_arg_ptr argv, int 439 static int count(struct user_arg_ptr argv, int max)
445 { 440 {
446 int i = 0; 441 int i = 0;
447 442
448 if (argv.ptr.native != NULL) { 443 if (argv.ptr.native != NULL) {
449 for (;;) { 444 for (;;) {
450 const char __user *p = 445 const char __user *p = get_user_arg_ptr(argv, i);
451 446
452 if (!p) 447 if (!p)
453 break; 448 break;
454 449
455 if (IS_ERR(p)) 450 if (IS_ERR(p))
456 return -EFAULT 451 return -EFAULT;
457 452
458 if (i >= max) 453 if (i >= max)
459 return -E2BIG; 454 return -E2BIG;
460 ++i; 455 ++i;
461 456
462 if (fatal_signal_pendi 457 if (fatal_signal_pending(current))
463 return -ERESTA 458 return -ERESTARTNOHAND;
464 cond_resched(); 459 cond_resched();
465 } 460 }
466 } 461 }
467 return i; 462 return i;
468 } 463 }
469 464
470 static int count_strings_kernel(const char *co <<
471 { <<
472 int i; <<
473 <<
474 if (!argv) <<
475 return 0; <<
476 <<
477 for (i = 0; argv[i]; ++i) { <<
478 if (i >= MAX_ARG_STRINGS) <<
479 return -E2BIG; <<
480 if (fatal_signal_pending(curre <<
481 return -ERESTARTNOHAND <<
482 cond_resched(); <<
483 } <<
484 return i; <<
485 } <<
486 <<
487 static inline int bprm_set_stack_limit(struct <<
488 unsigne <<
489 { <<
490 #ifdef CONFIG_MMU <<
491 /* Avoid a pathological bprm->p. */ <<
492 if (bprm->p < limit) <<
493 return -E2BIG; <<
494 bprm->argmin = bprm->p - limit; <<
495 #endif <<
496 return 0; <<
497 } <<
498 static inline bool bprm_hit_stack_limit(struct <<
499 { <<
500 #ifdef CONFIG_MMU <<
501 return bprm->p < bprm->argmin; <<
502 #else <<
503 return false; <<
504 #endif <<
505 } <<
506 <<
507 /* <<
508 * Calculate bprm->argmin from: <<
509 * - _STK_LIM <<
510 * - ARG_MAX <<
511 * - bprm->rlim_stack.rlim_cur <<
512 * - bprm->argc <<
513 * - bprm->envc <<
514 * - bprm->p <<
515 */ <<
516 static int bprm_stack_limits(struct linux_binp <<
517 { <<
518 unsigned long limit, ptr_size; <<
519 <<
520 /* <<
521 * Limit to 1/4 of the max stack size <<
522 * (whichever is smaller) for the argv <<
523 * This ensures that: <<
524 * - the remaining binfmt code will n <<
525 * - the program will have a reasonab <<
526 * to work from. <<
527 */ <<
528 limit = _STK_LIM / 4 * 3; <<
529 limit = min(limit, bprm->rlim_stack.rl <<
530 /* <<
531 * We've historically supported up to <<
532 * of argument strings even with small <<
533 */ <<
534 limit = max_t(unsigned long, limit, AR <<
535 /* Reject totally pathological counts. <<
536 if (bprm->argc < 0 || bprm->envc < 0) <<
537 return -E2BIG; <<
538 /* <<
539 * We must account for the size of all <<
540 * the argv and envp strings, since th <<
541 * the stack. They aren't stored until <<
542 * signal to the parent that the child <<
543 * Instead, calculate it here so it's <<
544 * <<
545 * In the case of argc = 0, make sure <<
546 * empty string (which will bump argc <<
547 * userspace programs don't start proc <<
548 * argc can never be 0, to keep them f <<
549 * See do_execveat_common(). <<
550 */ <<
551 if (check_add_overflow(max(bprm->argc, <<
552 check_mul_overflow(ptr_size, sizeo <<
553 return -E2BIG; <<
554 if (limit <= ptr_size) <<
555 return -E2BIG; <<
556 limit -= ptr_size; <<
557 <<
558 return bprm_set_stack_limit(bprm, limi <<
559 } <<
560 <<
561 /* 465 /*
562 * 'copy_strings()' copies argument/environmen 466 * 'copy_strings()' copies argument/environment strings from the old
563 * processes's memory to the new process's sta 467 * processes's memory to the new process's stack. The call to get_user_pages()
564 * ensures the destination page is created and 468 * ensures the destination page is created and not swapped out.
565 */ 469 */
566 static int copy_strings(int argc, struct user_ 470 static int copy_strings(int argc, struct user_arg_ptr argv,
567 struct linux_binprm *b 471 struct linux_binprm *bprm)
568 { 472 {
569 struct page *kmapped_page = NULL; 473 struct page *kmapped_page = NULL;
570 char *kaddr = NULL; 474 char *kaddr = NULL;
571 unsigned long kpos = 0; 475 unsigned long kpos = 0;
572 int ret; 476 int ret;
573 477
574 while (argc-- > 0) { 478 while (argc-- > 0) {
575 const char __user *str; 479 const char __user *str;
576 int len; 480 int len;
577 unsigned long pos; 481 unsigned long pos;
578 482
579 ret = -EFAULT; 483 ret = -EFAULT;
580 str = get_user_arg_ptr(argv, a 484 str = get_user_arg_ptr(argv, argc);
581 if (IS_ERR(str)) 485 if (IS_ERR(str))
582 goto out; 486 goto out;
583 487
584 len = strnlen_user(str, MAX_AR 488 len = strnlen_user(str, MAX_ARG_STRLEN);
585 if (!len) 489 if (!len)
586 goto out; 490 goto out;
587 491
588 ret = -E2BIG; 492 ret = -E2BIG;
589 if (!valid_arg_len(bprm, len)) 493 if (!valid_arg_len(bprm, len))
590 goto out; 494 goto out;
591 495
592 /* We're going to work our way !! 496 /* We're going to work our way backwords. */
593 pos = bprm->p; 497 pos = bprm->p;
594 str += len; 498 str += len;
595 bprm->p -= len; 499 bprm->p -= len;
596 if (bprm_hit_stack_limit(bprm) <<
597 goto out; <<
598 500
599 while (len > 0) { 501 while (len > 0) {
600 int offset, bytes_to_c 502 int offset, bytes_to_copy;
601 503
602 if (fatal_signal_pendi 504 if (fatal_signal_pending(current)) {
603 ret = -ERESTAR 505 ret = -ERESTARTNOHAND;
604 goto out; 506 goto out;
605 } 507 }
606 cond_resched(); 508 cond_resched();
607 509
608 offset = pos % PAGE_SI 510 offset = pos % PAGE_SIZE;
609 if (offset == 0) 511 if (offset == 0)
610 offset = PAGE_ 512 offset = PAGE_SIZE;
611 513
612 bytes_to_copy = offset 514 bytes_to_copy = offset;
613 if (bytes_to_copy > le 515 if (bytes_to_copy > len)
614 bytes_to_copy 516 bytes_to_copy = len;
615 517
616 offset -= bytes_to_cop 518 offset -= bytes_to_copy;
617 pos -= bytes_to_copy; 519 pos -= bytes_to_copy;
618 str -= bytes_to_copy; 520 str -= bytes_to_copy;
619 len -= bytes_to_copy; 521 len -= bytes_to_copy;
620 522
621 if (!kmapped_page || k 523 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
622 struct page *p 524 struct page *page;
623 525
624 page = get_arg 526 page = get_arg_page(bprm, pos, 1);
625 if (!page) { 527 if (!page) {
626 ret = 528 ret = -E2BIG;
627 goto o 529 goto out;
628 } 530 }
629 531
630 if (kmapped_pa 532 if (kmapped_page) {
631 flush_ !! 533 flush_kernel_dcache_page(kmapped_page);
632 kunmap !! 534 kunmap(kmapped_page);
633 put_ar 535 put_arg_page(kmapped_page);
634 } 536 }
635 kmapped_page = 537 kmapped_page = page;
636 kaddr = kmap_l !! 538 kaddr = kmap(kmapped_page);
637 kpos = pos & P 539 kpos = pos & PAGE_MASK;
638 flush_arg_page 540 flush_arg_page(bprm, kpos, kmapped_page);
639 } 541 }
640 if (copy_from_user(kad 542 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
641 ret = -EFAULT; 543 ret = -EFAULT;
642 goto out; 544 goto out;
643 } 545 }
644 } 546 }
645 } 547 }
646 ret = 0; 548 ret = 0;
647 out: 549 out:
648 if (kmapped_page) { 550 if (kmapped_page) {
649 flush_dcache_page(kmapped_page !! 551 flush_kernel_dcache_page(kmapped_page);
650 kunmap_local(kaddr); !! 552 kunmap(kmapped_page);
651 put_arg_page(kmapped_page); 553 put_arg_page(kmapped_page);
652 } 554 }
653 return ret; 555 return ret;
654 } 556 }
655 557
656 /* 558 /*
657 * Copy and argument/environment string from t !! 559 * Like copy_strings, but get argv and its values from kernel memory.
658 */ 560 */
659 int copy_string_kernel(const char *arg, struct !! 561 int copy_strings_kernel(int argc, const char *const *__argv,
>> 562 struct linux_binprm *bprm)
660 { 563 {
661 int len = strnlen(arg, MAX_ARG_STRLEN) !! 564 int r;
662 unsigned long pos = bprm->p; !! 565 mm_segment_t oldfs = get_fs();
>> 566 struct user_arg_ptr argv = {
>> 567 .ptr.native = (const char __user *const __user *)__argv,
>> 568 };
663 569
664 if (len == 0) !! 570 set_fs(KERNEL_DS);
>> 571 r = copy_strings(argc, argv, bprm);
>> 572 set_fs(oldfs);
>> 573
>> 574 return r;
>> 575 }
>> 576 EXPORT_SYMBOL(copy_strings_kernel);
>> 577
>> 578 #ifdef CONFIG_MMU
>> 579
>> 580 /*
>> 581 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
>> 582 * the binfmt code determines where the new stack should reside, we shift it to
>> 583 * its final location. The process proceeds as follows:
>> 584 *
>> 585 * 1) Use shift to calculate the new vma endpoints.
>> 586 * 2) Extend vma to cover both the old and new ranges. This ensures the
>> 587 * arguments passed to subsequent functions are consistent.
>> 588 * 3) Move vma's page tables to the new range.
>> 589 * 4) Free up any cleared pgd range.
>> 590 * 5) Shrink the vma to cover only the new range.
>> 591 */
>> 592 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
>> 593 {
>> 594 struct mm_struct *mm = vma->vm_mm;
>> 595 unsigned long old_start = vma->vm_start;
>> 596 unsigned long old_end = vma->vm_end;
>> 597 unsigned long length = old_end - old_start;
>> 598 unsigned long new_start = old_start - shift;
>> 599 unsigned long new_end = old_end - shift;
>> 600 struct mmu_gather tlb;
>> 601
>> 602 BUG_ON(new_start > new_end);
>> 603
>> 604 /*
>> 605 * ensure there are no vmas between where we want to go
>> 606 * and where we are
>> 607 */
>> 608 if (vma != find_vma(mm, new_start))
665 return -EFAULT; 609 return -EFAULT;
666 if (!valid_arg_len(bprm, len)) <<
667 return -E2BIG; <<
668 610
669 /* We're going to work our way backwar !! 611 /*
670 arg += len; !! 612 * cover the whole range: [new_start, old_end)
671 bprm->p -= len; !! 613 */
672 if (bprm_hit_stack_limit(bprm)) !! 614 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
673 return -E2BIG; !! 615 return -ENOMEM;
674 <<
675 while (len > 0) { <<
676 unsigned int bytes_to_copy = m <<
677 min_not_zero(o <<
678 struct page *page; <<
679 <<
680 pos -= bytes_to_copy; <<
681 arg -= bytes_to_copy; <<
682 len -= bytes_to_copy; <<
683 616
684 page = get_arg_page(bprm, pos, !! 617 /*
685 if (!page) !! 618 * move the page tables downwards, on failure we rely on
686 return -E2BIG; !! 619 * process cleanup to remove whatever mess we made.
687 flush_arg_page(bprm, pos & PAG !! 620 */
688 memcpy_to_page(page, offset_in !! 621 if (length != move_page_tables(vma, old_start,
689 put_arg_page(page); !! 622 vma, new_start, length, false))
>> 623 return -ENOMEM;
>> 624
>> 625 lru_add_drain();
>> 626 tlb_gather_mmu(&tlb, mm, old_start, old_end);
>> 627 if (new_end > old_start) {
>> 628 /*
>> 629 * when the old and new regions overlap clear from new_end.
>> 630 */
>> 631 free_pgd_range(&tlb, new_end, old_end, new_end,
>> 632 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
>> 633 } else {
>> 634 /*
>> 635 * otherwise, clean from old_start; this is done to not touch
>> 636 * the address space in [new_end, old_start) some architectures
>> 637 * have constraints on va-space that make this illegal (IA64) -
>> 638 * for the others its just a little faster.
>> 639 */
>> 640 free_pgd_range(&tlb, old_start, old_end, new_end,
>> 641 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
690 } 642 }
>> 643 tlb_finish_mmu(&tlb, old_start, old_end);
691 644
692 return 0; !! 645 /*
693 } !! 646 * Shrink the vma to just the new range. Always succeeds.
694 EXPORT_SYMBOL(copy_string_kernel); !! 647 */
>> 648 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
695 649
696 static int copy_strings_kernel(int argc, const <<
697 struct linux_bi <<
698 { <<
699 while (argc-- > 0) { <<
700 int ret = copy_string_kernel(a <<
701 if (ret < 0) <<
702 return ret; <<
703 if (fatal_signal_pending(curre <<
704 return -ERESTARTNOHAND <<
705 cond_resched(); <<
706 } <<
707 return 0; 650 return 0;
708 } 651 }
709 652
710 #ifdef CONFIG_MMU <<
711 <<
712 /* 653 /*
713 * Finalizes the stack vm_area_struct. The fla 654 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
714 * the stack is optionally relocated, and some 655 * the stack is optionally relocated, and some extra space is added.
715 */ 656 */
716 int setup_arg_pages(struct linux_binprm *bprm, 657 int setup_arg_pages(struct linux_binprm *bprm,
717 unsigned long stack_top, 658 unsigned long stack_top,
718 int executable_stack) 659 int executable_stack)
719 { 660 {
720 unsigned long ret; 661 unsigned long ret;
721 unsigned long stack_shift; 662 unsigned long stack_shift;
722 struct mm_struct *mm = current->mm; 663 struct mm_struct *mm = current->mm;
723 struct vm_area_struct *vma = bprm->vma 664 struct vm_area_struct *vma = bprm->vma;
724 struct vm_area_struct *prev = NULL; 665 struct vm_area_struct *prev = NULL;
725 unsigned long vm_flags; 666 unsigned long vm_flags;
726 unsigned long stack_base; 667 unsigned long stack_base;
727 unsigned long stack_size; 668 unsigned long stack_size;
728 unsigned long stack_expand; 669 unsigned long stack_expand;
729 unsigned long rlim_stack; 670 unsigned long rlim_stack;
730 struct mmu_gather tlb; <<
731 struct vma_iterator vmi; <<
732 671
733 #ifdef CONFIG_STACK_GROWSUP 672 #ifdef CONFIG_STACK_GROWSUP
734 /* Limit stack size */ 673 /* Limit stack size */
735 stack_base = bprm->rlim_stack.rlim_max !! 674 stack_base = rlimit_max(RLIMIT_STACK);
736 !! 675 if (stack_base > STACK_SIZE_MAX)
737 stack_base = calc_max_stack_size(stack !! 676 stack_base = STACK_SIZE_MAX;
738 677
739 /* Add space for stack randomization. 678 /* Add space for stack randomization. */
740 if (current->flags & PF_RANDOMIZE) !! 679 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
741 stack_base += (STACK_RND_MASK <<
742 680
743 /* Make sure we didn't let the argumen 681 /* Make sure we didn't let the argument array grow too large. */
744 if (vma->vm_end - vma->vm_start > stac 682 if (vma->vm_end - vma->vm_start > stack_base)
745 return -ENOMEM; 683 return -ENOMEM;
746 684
747 stack_base = PAGE_ALIGN(stack_top - st 685 stack_base = PAGE_ALIGN(stack_top - stack_base);
748 686
749 stack_shift = vma->vm_start - stack_ba 687 stack_shift = vma->vm_start - stack_base;
750 mm->arg_start = bprm->p - stack_shift; 688 mm->arg_start = bprm->p - stack_shift;
751 bprm->p = vma->vm_end - stack_shift; 689 bprm->p = vma->vm_end - stack_shift;
752 #else 690 #else
753 stack_top = arch_align_stack(stack_top 691 stack_top = arch_align_stack(stack_top);
754 stack_top = PAGE_ALIGN(stack_top); 692 stack_top = PAGE_ALIGN(stack_top);
755 693
756 if (unlikely(stack_top < mmap_min_addr 694 if (unlikely(stack_top < mmap_min_addr) ||
757 unlikely(vma->vm_end - vma->vm_sta 695 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
758 return -ENOMEM; 696 return -ENOMEM;
759 697
760 stack_shift = vma->vm_end - stack_top; 698 stack_shift = vma->vm_end - stack_top;
761 699
762 bprm->p -= stack_shift; 700 bprm->p -= stack_shift;
763 mm->arg_start = bprm->p; 701 mm->arg_start = bprm->p;
764 #endif 702 #endif
765 703
766 if (bprm->loader) 704 if (bprm->loader)
767 bprm->loader -= stack_shift; 705 bprm->loader -= stack_shift;
768 bprm->exec -= stack_shift; 706 bprm->exec -= stack_shift;
769 707
770 if (mmap_write_lock_killable(mm)) !! 708 if (down_write_killable(&mm->mmap_sem))
771 return -EINTR; 709 return -EINTR;
772 710
773 vm_flags = VM_STACK_FLAGS; 711 vm_flags = VM_STACK_FLAGS;
774 712
775 /* 713 /*
776 * Adjust stack execute permissions; e 714 * Adjust stack execute permissions; explicitly enable for
777 * EXSTACK_ENABLE_X, disable for EXSTA 715 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
778 * (arch default) otherwise. 716 * (arch default) otherwise.
779 */ 717 */
780 if (unlikely(executable_stack == EXSTA 718 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
781 vm_flags |= VM_EXEC; 719 vm_flags |= VM_EXEC;
782 else if (executable_stack == EXSTACK_D 720 else if (executable_stack == EXSTACK_DISABLE_X)
783 vm_flags &= ~VM_EXEC; 721 vm_flags &= ~VM_EXEC;
784 vm_flags |= mm->def_flags; 722 vm_flags |= mm->def_flags;
785 vm_flags |= VM_STACK_INCOMPLETE_SETUP; 723 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
786 724
787 vma_iter_init(&vmi, mm, vma->vm_start) !! 725 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
788 <<
789 tlb_gather_mmu(&tlb, mm); <<
790 ret = mprotect_fixup(&vmi, &tlb, vma, <<
791 vm_flags); 726 vm_flags);
792 tlb_finish_mmu(&tlb); <<
793 <<
794 if (ret) 727 if (ret)
795 goto out_unlock; 728 goto out_unlock;
796 BUG_ON(prev != vma); 729 BUG_ON(prev != vma);
797 730
798 if (unlikely(vm_flags & VM_EXEC)) { <<
799 pr_warn_once("process '%pD4' s <<
800 bprm->file); <<
801 } <<
802 <<
803 /* Move stack pages down in memory. */ 731 /* Move stack pages down in memory. */
804 if (stack_shift) { 732 if (stack_shift) {
805 /* !! 733 ret = shift_arg_pages(vma, stack_shift);
806 * During bprm_mm_init(), we c <<
807 * the binfmt code determines <<
808 * its final location. <<
809 */ <<
810 ret = relocate_vma_down(vma, s <<
811 if (ret) 734 if (ret)
812 goto out_unlock; 735 goto out_unlock;
813 } 736 }
814 737
815 /* mprotect_fixup is overkill to remov 738 /* mprotect_fixup is overkill to remove the temporary stack flags */
816 vm_flags_clear(vma, VM_STACK_INCOMPLET !! 739 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
817 740
818 stack_expand = 131072UL; /* randomly 3 741 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
819 stack_size = vma->vm_end - vma->vm_sta 742 stack_size = vma->vm_end - vma->vm_start;
820 /* 743 /*
821 * Align this down to a page boundary 744 * Align this down to a page boundary as expand_stack
822 * will align it up. 745 * will align it up.
823 */ 746 */
824 rlim_stack = bprm->rlim_stack.rlim_cur !! 747 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
825 <<
826 stack_expand = min(rlim_stack, stack_s <<
827 <<
828 #ifdef CONFIG_STACK_GROWSUP 748 #ifdef CONFIG_STACK_GROWSUP
829 stack_base = vma->vm_start + stack_exp !! 749 if (stack_size + stack_expand > rlim_stack)
>> 750 stack_base = vma->vm_start + rlim_stack;
>> 751 else
>> 752 stack_base = vma->vm_end + stack_expand;
830 #else 753 #else
831 stack_base = vma->vm_end - stack_expan !! 754 if (stack_size + stack_expand > rlim_stack)
>> 755 stack_base = vma->vm_end - rlim_stack;
>> 756 else
>> 757 stack_base = vma->vm_start - stack_expand;
832 #endif 758 #endif
833 current->mm->start_stack = bprm->p; 759 current->mm->start_stack = bprm->p;
834 ret = expand_stack_locked(vma, stack_b !! 760 ret = expand_stack(vma, stack_base);
835 if (ret) 761 if (ret)
836 ret = -EFAULT; 762 ret = -EFAULT;
837 763
838 out_unlock: 764 out_unlock:
839 mmap_write_unlock(mm); !! 765 up_write(&mm->mmap_sem);
840 return ret; 766 return ret;
841 } 767 }
842 EXPORT_SYMBOL(setup_arg_pages); 768 EXPORT_SYMBOL(setup_arg_pages);
843 769
844 #else 770 #else
845 771
846 /* 772 /*
847 * Transfer the program arguments and environm 773 * Transfer the program arguments and environment from the holding pages
848 * onto the stack. The provided stack pointer 774 * onto the stack. The provided stack pointer is adjusted accordingly.
849 */ 775 */
850 int transfer_args_to_stack(struct linux_binprm 776 int transfer_args_to_stack(struct linux_binprm *bprm,
851 unsigned long *sp_l 777 unsigned long *sp_location)
852 { 778 {
853 unsigned long index, stop, sp; 779 unsigned long index, stop, sp;
854 int ret = 0; 780 int ret = 0;
855 781
856 stop = bprm->p >> PAGE_SHIFT; 782 stop = bprm->p >> PAGE_SHIFT;
857 sp = *sp_location; 783 sp = *sp_location;
858 784
859 for (index = MAX_ARG_PAGES - 1; index 785 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
860 unsigned int offset = index == 786 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
861 char *src = kmap_local_page(bp !! 787 char *src = kmap(bprm->page[index]) + offset;
862 sp -= PAGE_SIZE - offset; 788 sp -= PAGE_SIZE - offset;
863 if (copy_to_user((void *) sp, 789 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
864 ret = -EFAULT; 790 ret = -EFAULT;
865 kunmap_local(src); !! 791 kunmap(bprm->page[index]);
866 if (ret) 792 if (ret)
867 goto out; 793 goto out;
868 } 794 }
869 795
870 bprm->exec += *sp_location - MAX_ARG_P <<
871 *sp_location = sp; 796 *sp_location = sp;
872 797
873 out: 798 out:
874 return ret; 799 return ret;
875 } 800 }
876 EXPORT_SYMBOL(transfer_args_to_stack); 801 EXPORT_SYMBOL(transfer_args_to_stack);
877 802
878 #endif /* CONFIG_MMU */ 803 #endif /* CONFIG_MMU */
879 804
880 /* <<
881 * On success, caller must call do_close_execa <<
882 * struct file to close it. <<
883 */ <<
884 static struct file *do_open_execat(int fd, str 805 static struct file *do_open_execat(int fd, struct filename *name, int flags)
885 { 806 {
886 struct file *file; 807 struct file *file;
>> 808 int err;
887 struct open_flags open_exec_flags = { 809 struct open_flags open_exec_flags = {
888 .open_flag = O_LARGEFILE | O_R 810 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
889 .acc_mode = MAY_EXEC, 811 .acc_mode = MAY_EXEC,
890 .intent = LOOKUP_OPEN, 812 .intent = LOOKUP_OPEN,
891 .lookup_flags = LOOKUP_FOLLOW, 813 .lookup_flags = LOOKUP_FOLLOW,
892 }; 814 };
893 815
894 if ((flags & ~(AT_SYMLINK_NOFOLLOW | A 816 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
895 return ERR_PTR(-EINVAL); 817 return ERR_PTR(-EINVAL);
896 if (flags & AT_SYMLINK_NOFOLLOW) 818 if (flags & AT_SYMLINK_NOFOLLOW)
897 open_exec_flags.lookup_flags & 819 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
898 if (flags & AT_EMPTY_PATH) 820 if (flags & AT_EMPTY_PATH)
899 open_exec_flags.lookup_flags | 821 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
900 822
901 file = do_filp_open(fd, name, &open_ex 823 file = do_filp_open(fd, name, &open_exec_flags);
902 if (IS_ERR(file)) 824 if (IS_ERR(file))
903 return file; !! 825 goto out;
904 826
905 /* !! 827 err = -EACCES;
906 * In the past the regular type check !! 828 if (!S_ISREG(file_inode(file)->i_mode))
907 * 633fb6ac3980 ("exec: move S_ISREG() !! 829 goto exit;
908 * an invariant that all non-regular f !! 830
909 */ !! 831 if (path_noexec(&file->f_path))
910 if (WARN_ON_ONCE(!S_ISREG(file_inode(f !! 832 goto exit;
911 path_noexec(&file->f_path)) { <<
912 fput(file); <<
913 return ERR_PTR(-EACCES); <<
914 } <<
915 833
>> 834 err = deny_write_access(file);
>> 835 if (err)
>> 836 goto exit;
>> 837
>> 838 if (name->name[0] != '\0')
>> 839 fsnotify_open(file);
>> 840
>> 841 out:
916 return file; 842 return file;
>> 843
>> 844 exit:
>> 845 fput(file);
>> 846 return ERR_PTR(err);
917 } 847 }
918 848
919 /** <<
920 * open_exec - Open a path name for execution <<
921 * <<
922 * @name: path name to open with the intent of <<
923 * <<
924 * Returns ERR_PTR on failure or allocated str <<
925 * <<
926 * As this is a wrapper for the internal do_op <<
927 * do_close_execat(). <<
928 */ <<
929 struct file *open_exec(const char *name) 849 struct file *open_exec(const char *name)
930 { 850 {
931 struct filename *filename = getname_ke 851 struct filename *filename = getname_kernel(name);
932 struct file *f = ERR_CAST(filename); 852 struct file *f = ERR_CAST(filename);
933 853
934 if (!IS_ERR(filename)) { 854 if (!IS_ERR(filename)) {
935 f = do_open_execat(AT_FDCWD, f 855 f = do_open_execat(AT_FDCWD, filename, 0);
936 putname(filename); 856 putname(filename);
937 } 857 }
938 return f; 858 return f;
939 } 859 }
940 EXPORT_SYMBOL(open_exec); 860 EXPORT_SYMBOL(open_exec);
941 861
942 #if defined(CONFIG_BINFMT_FLAT) || defined(CON !! 862 int kernel_read(struct file *file, loff_t offset,
>> 863 char *addr, unsigned long count)
>> 864 {
>> 865 mm_segment_t old_fs;
>> 866 loff_t pos = offset;
>> 867 int result;
>> 868
>> 869 old_fs = get_fs();
>> 870 set_fs(get_ds());
>> 871 /* The cast to a user pointer is valid due to the set_fs() */
>> 872 result = vfs_read(file, (void __user *)addr, count, &pos);
>> 873 set_fs(old_fs);
>> 874 return result;
>> 875 }
>> 876
>> 877 EXPORT_SYMBOL(kernel_read);
>> 878
>> 879 int kernel_read_file(struct file *file, void **buf, loff_t *size,
>> 880 loff_t max_size, enum kernel_read_file_id id)
>> 881 {
>> 882 loff_t i_size, pos;
>> 883 ssize_t bytes = 0;
>> 884 int ret;
>> 885
>> 886 if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
>> 887 return -EINVAL;
>> 888
>> 889 ret = security_kernel_read_file(file, id);
>> 890 if (ret)
>> 891 return ret;
>> 892
>> 893 ret = deny_write_access(file);
>> 894 if (ret)
>> 895 return ret;
>> 896
>> 897 i_size = i_size_read(file_inode(file));
>> 898 if (max_size > 0 && i_size > max_size) {
>> 899 ret = -EFBIG;
>> 900 goto out;
>> 901 }
>> 902 if (i_size <= 0) {
>> 903 ret = -EINVAL;
>> 904 goto out;
>> 905 }
>> 906
>> 907 if (id != READING_FIRMWARE_PREALLOC_BUFFER)
>> 908 *buf = vmalloc(i_size);
>> 909 if (!*buf) {
>> 910 ret = -ENOMEM;
>> 911 goto out;
>> 912 }
>> 913
>> 914 pos = 0;
>> 915 while (pos < i_size) {
>> 916 bytes = kernel_read(file, pos, (char *)(*buf) + pos,
>> 917 i_size - pos);
>> 918 if (bytes < 0) {
>> 919 ret = bytes;
>> 920 goto out;
>> 921 }
>> 922
>> 923 if (bytes == 0)
>> 924 break;
>> 925 pos += bytes;
>> 926 }
>> 927
>> 928 if (pos != i_size) {
>> 929 ret = -EIO;
>> 930 goto out_free;
>> 931 }
>> 932
>> 933 ret = security_kernel_post_read_file(file, *buf, i_size, id);
>> 934 if (!ret)
>> 935 *size = pos;
>> 936
>> 937 out_free:
>> 938 if (ret < 0) {
>> 939 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
>> 940 vfree(*buf);
>> 941 *buf = NULL;
>> 942 }
>> 943 }
>> 944
>> 945 out:
>> 946 allow_write_access(file);
>> 947 return ret;
>> 948 }
>> 949 EXPORT_SYMBOL_GPL(kernel_read_file);
>> 950
>> 951 int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
>> 952 loff_t max_size, enum kernel_read_file_id id)
>> 953 {
>> 954 struct file *file;
>> 955 int ret;
>> 956
>> 957 if (!path || !*path)
>> 958 return -EINVAL;
>> 959
>> 960 file = filp_open(path, O_RDONLY, 0);
>> 961 if (IS_ERR(file))
>> 962 return PTR_ERR(file);
>> 963
>> 964 ret = kernel_read_file(file, buf, size, max_size, id);
>> 965 fput(file);
>> 966 return ret;
>> 967 }
>> 968 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
>> 969
>> 970 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
>> 971 enum kernel_read_file_id id)
>> 972 {
>> 973 struct fd f = fdget(fd);
>> 974 int ret = -EBADF;
>> 975
>> 976 if (!f.file)
>> 977 goto out;
>> 978
>> 979 ret = kernel_read_file(f.file, buf, size, max_size, id);
>> 980 out:
>> 981 fdput(f);
>> 982 return ret;
>> 983 }
>> 984 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
>> 985
943 ssize_t read_code(struct file *file, unsigned 986 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
944 { 987 {
945 ssize_t res = vfs_read(file, (void __u 988 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
946 if (res > 0) 989 if (res > 0)
947 flush_icache_user_range(addr, !! 990 flush_icache_range(addr, addr + len);
948 return res; 991 return res;
949 } 992 }
950 EXPORT_SYMBOL(read_code); 993 EXPORT_SYMBOL(read_code);
951 #endif <<
952 994
953 /* <<
954 * Maps the mm_struct mm into the current task <<
955 * On success, this function returns with exec <<
956 * held for writing. <<
957 */ <<
958 static int exec_mmap(struct mm_struct *mm) 995 static int exec_mmap(struct mm_struct *mm)
959 { 996 {
960 struct task_struct *tsk; 997 struct task_struct *tsk;
961 struct mm_struct *old_mm, *active_mm; 998 struct mm_struct *old_mm, *active_mm;
962 int ret; <<
963 999
964 /* Notify parent that we're no longer 1000 /* Notify parent that we're no longer interested in the old VM */
965 tsk = current; 1001 tsk = current;
966 old_mm = current->mm; 1002 old_mm = current->mm;
967 exec_mm_release(tsk, old_mm); !! 1003 mm_release(tsk, old_mm);
968 <<
969 ret = down_write_killable(&tsk->signal <<
970 if (ret) <<
971 return ret; <<
972 1004
973 if (old_mm) { 1005 if (old_mm) {
>> 1006 sync_mm_rss(old_mm);
974 /* 1007 /*
975 * If there is a pending fatal !! 1008 * Make sure that if there is a core dump in progress
976 * whose default action is to !! 1009 * for the old mm, we get out and die instead of going
977 * out and die instead of goin !! 1010 * through with the exec. We must hold mmap_sem around
>> 1011 * checking core_state and changing tsk->mm.
978 */ 1012 */
979 ret = mmap_read_lock_killable( !! 1013 down_read(&old_mm->mmap_sem);
980 if (ret) { !! 1014 if (unlikely(old_mm->core_state)) {
981 up_write(&tsk->signal- !! 1015 up_read(&old_mm->mmap_sem);
982 return ret; !! 1016 return -EINTR;
983 } 1017 }
984 } 1018 }
985 <<
986 task_lock(tsk); 1019 task_lock(tsk);
987 membarrier_exec_mmap(mm); <<
988 <<
989 local_irq_disable(); <<
990 active_mm = tsk->active_mm; 1020 active_mm = tsk->active_mm;
991 tsk->active_mm = mm; <<
992 tsk->mm = mm; 1021 tsk->mm = mm;
993 mm_init_cid(mm); !! 1022 tsk->active_mm = mm;
994 /* <<
995 * This prevents preemption while acti <<
996 * it and mm are being updated, which <<
997 * lazy tlb mm refcounting when these <<
998 * switches. Not all architectures can <<
999 * activate_mm yet. <<
1000 */ <<
1001 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS <<
1002 local_irq_enable(); <<
1003 activate_mm(active_mm, mm); 1023 activate_mm(active_mm, mm);
1004 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_ !! 1024 tsk->mm->vmacache_seqnum = 0;
1005 local_irq_enable(); !! 1025 vmacache_flush(tsk);
1006 lru_gen_add_mm(mm); <<
1007 task_unlock(tsk); 1026 task_unlock(tsk);
1008 lru_gen_use_mm(mm); <<
1009 if (old_mm) { 1027 if (old_mm) {
1010 mmap_read_unlock(old_mm); !! 1028 up_read(&old_mm->mmap_sem);
1011 BUG_ON(active_mm != old_mm); 1029 BUG_ON(active_mm != old_mm);
1012 setmax_mm_hiwater_rss(&tsk->s 1030 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1013 mm_update_next_owner(old_mm); 1031 mm_update_next_owner(old_mm);
1014 mmput(old_mm); 1032 mmput(old_mm);
1015 return 0; 1033 return 0;
1016 } 1034 }
1017 mmdrop_lazy_tlb(active_mm); !! 1035 mmdrop(active_mm);
1018 return 0; 1036 return 0;
1019 } 1037 }
1020 1038
>> 1039 /*
>> 1040 * This function makes sure the current process has its own signal table,
>> 1041 * so that flush_signal_handlers can later reset the handlers without
>> 1042 * disturbing other processes. (Other processes might share the signal
>> 1043 * table via the CLONE_SIGHAND option to clone().)
>> 1044 */
1021 static int de_thread(struct task_struct *tsk) 1045 static int de_thread(struct task_struct *tsk)
1022 { 1046 {
1023 struct signal_struct *sig = tsk->sign 1047 struct signal_struct *sig = tsk->signal;
1024 struct sighand_struct *oldsighand = t 1048 struct sighand_struct *oldsighand = tsk->sighand;
1025 spinlock_t *lock = &oldsighand->siglo 1049 spinlock_t *lock = &oldsighand->siglock;
1026 1050
1027 if (thread_group_empty(tsk)) 1051 if (thread_group_empty(tsk))
1028 goto no_thread_group; 1052 goto no_thread_group;
1029 1053
1030 /* 1054 /*
1031 * Kill all other threads in the thre 1055 * Kill all other threads in the thread group.
1032 */ 1056 */
1033 spin_lock_irq(lock); 1057 spin_lock_irq(lock);
1034 if ((sig->flags & SIGNAL_GROUP_EXIT) !! 1058 if (signal_group_exit(sig)) {
1035 /* 1059 /*
1036 * Another group action in pr 1060 * Another group action in progress, just
1037 * return so that the signal 1061 * return so that the signal is processed.
1038 */ 1062 */
1039 spin_unlock_irq(lock); 1063 spin_unlock_irq(lock);
1040 return -EAGAIN; 1064 return -EAGAIN;
1041 } 1065 }
1042 1066
1043 sig->group_exec_task = tsk; !! 1067 sig->group_exit_task = tsk;
1044 sig->notify_count = zap_other_threads 1068 sig->notify_count = zap_other_threads(tsk);
1045 if (!thread_group_leader(tsk)) 1069 if (!thread_group_leader(tsk))
1046 sig->notify_count--; 1070 sig->notify_count--;
1047 1071
1048 while (sig->notify_count) { 1072 while (sig->notify_count) {
1049 __set_current_state(TASK_KILL 1073 __set_current_state(TASK_KILLABLE);
1050 spin_unlock_irq(lock); 1074 spin_unlock_irq(lock);
1051 schedule(); 1075 schedule();
1052 if (__fatal_signal_pending(ts !! 1076 if (unlikely(__fatal_signal_pending(tsk)))
1053 goto killed; 1077 goto killed;
1054 spin_lock_irq(lock); 1078 spin_lock_irq(lock);
1055 } 1079 }
1056 spin_unlock_irq(lock); 1080 spin_unlock_irq(lock);
1057 1081
1058 /* 1082 /*
1059 * At this point all other threads ha 1083 * At this point all other threads have exited, all we have to
1060 * do is to wait for the thread group 1084 * do is to wait for the thread group leader to become inactive,
1061 * and to assume its PID: 1085 * and to assume its PID:
1062 */ 1086 */
1063 if (!thread_group_leader(tsk)) { 1087 if (!thread_group_leader(tsk)) {
1064 struct task_struct *leader = 1088 struct task_struct *leader = tsk->group_leader;
1065 1089
1066 for (;;) { 1090 for (;;) {
1067 cgroup_threadgroup_ch !! 1091 threadgroup_change_begin(tsk);
1068 write_lock_irq(&taskl 1092 write_lock_irq(&tasklist_lock);
1069 /* 1093 /*
1070 * Do this under task 1094 * Do this under tasklist_lock to ensure that
1071 * exit_notify() can' !! 1095 * exit_notify() can't miss ->group_exit_task
1072 */ 1096 */
1073 sig->notify_count = - 1097 sig->notify_count = -1;
1074 if (likely(leader->ex 1098 if (likely(leader->exit_state))
1075 break; 1099 break;
1076 __set_current_state(T 1100 __set_current_state(TASK_KILLABLE);
1077 write_unlock_irq(&tas 1101 write_unlock_irq(&tasklist_lock);
1078 cgroup_threadgroup_ch !! 1102 threadgroup_change_end(tsk);
1079 schedule(); 1103 schedule();
1080 if (__fatal_signal_pe !! 1104 if (unlikely(__fatal_signal_pending(tsk)))
1081 goto killed; 1105 goto killed;
1082 } 1106 }
1083 1107
1084 /* 1108 /*
1085 * The only record we have of 1109 * The only record we have of the real-time age of a
1086 * process, regardless of exe 1110 * process, regardless of execs it's done, is start_time.
1087 * All the past CPU time is a 1111 * All the past CPU time is accumulated in signal_struct
1088 * from sister threads now de 1112 * from sister threads now dead. But in this non-leader
1089 * exec, nothing survives fro 1113 * exec, nothing survives from the original leader thread,
1090 * whose birth marks the true 1114 * whose birth marks the true age of this process now.
1091 * When we take on its identi 1115 * When we take on its identity by switching to its PID, we
1092 * also take its birthdate (a 1116 * also take its birthdate (always earlier than our own).
1093 */ 1117 */
1094 tsk->start_time = leader->sta 1118 tsk->start_time = leader->start_time;
1095 tsk->start_boottime = leader- !! 1119 tsk->real_start_time = leader->real_start_time;
1096 1120
1097 BUG_ON(!same_thread_group(lea 1121 BUG_ON(!same_thread_group(leader, tsk));
>> 1122 BUG_ON(has_group_leader_pid(tsk));
1098 /* 1123 /*
1099 * An exec() starts a new thr 1124 * An exec() starts a new thread group with the
1100 * TGID of the previous threa 1125 * TGID of the previous thread group. Rehash the
1101 * two threads with a switche 1126 * two threads with a switched PID, and release
1102 * the former thread group le 1127 * the former thread group leader:
1103 */ 1128 */
1104 1129
1105 /* Become a process group lea 1130 /* Become a process group leader with the old leader's pid.
1106 * The old leader becomes a t 1131 * The old leader becomes a thread of the this thread group.
>> 1132 * Note: The old leader also uses this pid until release_task
>> 1133 * is called. Odd but simple and correct.
1107 */ 1134 */
1108 exchange_tids(tsk, leader); !! 1135 tsk->pid = leader->pid;
1109 transfer_pid(leader, tsk, PID !! 1136 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1110 transfer_pid(leader, tsk, PID 1137 transfer_pid(leader, tsk, PIDTYPE_PGID);
1111 transfer_pid(leader, tsk, PID 1138 transfer_pid(leader, tsk, PIDTYPE_SID);
1112 1139
1113 list_replace_rcu(&leader->tas 1140 list_replace_rcu(&leader->tasks, &tsk->tasks);
1114 list_replace_init(&leader->si 1141 list_replace_init(&leader->sibling, &tsk->sibling);
1115 1142
1116 tsk->group_leader = tsk; 1143 tsk->group_leader = tsk;
1117 leader->group_leader = tsk; 1144 leader->group_leader = tsk;
1118 1145
1119 tsk->exit_signal = SIGCHLD; 1146 tsk->exit_signal = SIGCHLD;
1120 leader->exit_signal = -1; 1147 leader->exit_signal = -1;
1121 1148
1122 BUG_ON(leader->exit_state != 1149 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1123 leader->exit_state = EXIT_DEA 1150 leader->exit_state = EXIT_DEAD;
>> 1151
1124 /* 1152 /*
1125 * We are going to release_ta 1153 * We are going to release_task()->ptrace_unlink() silently,
1126 * the tracer can sleep in do 1154 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1127 * the tracer won't block aga !! 1155 * the tracer wont't block again waiting for this thread.
1128 */ 1156 */
1129 if (unlikely(leader->ptrace)) 1157 if (unlikely(leader->ptrace))
1130 __wake_up_parent(lead 1158 __wake_up_parent(leader, leader->parent);
1131 write_unlock_irq(&tasklist_lo 1159 write_unlock_irq(&tasklist_lock);
1132 cgroup_threadgroup_change_end !! 1160 threadgroup_change_end(tsk);
1133 1161
1134 release_task(leader); 1162 release_task(leader);
1135 } 1163 }
1136 1164
1137 sig->group_exec_task = NULL; !! 1165 sig->group_exit_task = NULL;
1138 sig->notify_count = 0; 1166 sig->notify_count = 0;
1139 1167
1140 no_thread_group: 1168 no_thread_group:
1141 /* we have changed execution domain * 1169 /* we have changed execution domain */
1142 tsk->exit_signal = SIGCHLD; 1170 tsk->exit_signal = SIGCHLD;
1143 1171
1144 BUG_ON(!thread_group_leader(tsk)); !! 1172 #ifdef CONFIG_POSIX_TIMERS
1145 return 0; !! 1173 exit_itimers(sig);
1146 !! 1174 flush_itimer_signals();
1147 killed: !! 1175 #endif
1148 /* protects against exit_notify() and <<
1149 read_lock(&tasklist_lock); <<
1150 sig->group_exec_task = NULL; <<
1151 sig->notify_count = 0; <<
1152 read_unlock(&tasklist_lock); <<
1153 return -EAGAIN; <<
1154 } <<
1155 <<
1156 <<
1157 /* <<
1158 * This function makes sure the current proce <<
1159 * so that flush_signal_handlers can later re <<
1160 * disturbing other processes. (Other proces <<
1161 * table via the CLONE_SIGHAND option to clon <<
1162 */ <<
1163 static int unshare_sighand(struct task_struct <<
1164 { <<
1165 struct sighand_struct *oldsighand = m <<
1166 1176
1167 if (refcount_read(&oldsighand->count) !! 1177 if (atomic_read(&oldsighand->count) != 1) {
1168 struct sighand_struct *newsig 1178 struct sighand_struct *newsighand;
1169 /* 1179 /*
1170 * This ->sighand is shared w 1180 * This ->sighand is shared with the CLONE_SIGHAND
1171 * but not CLONE_THREAD task, 1181 * but not CLONE_THREAD task, switch to the new one.
1172 */ 1182 */
1173 newsighand = kmem_cache_alloc 1183 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1174 if (!newsighand) 1184 if (!newsighand)
1175 return -ENOMEM; 1185 return -ENOMEM;
1176 1186
1177 refcount_set(&newsighand->cou !! 1187 atomic_set(&newsighand->count, 1);
>> 1188 memcpy(newsighand->action, oldsighand->action,
>> 1189 sizeof(newsighand->action));
1178 1190
1179 write_lock_irq(&tasklist_lock 1191 write_lock_irq(&tasklist_lock);
1180 spin_lock(&oldsighand->sigloc 1192 spin_lock(&oldsighand->siglock);
1181 memcpy(newsighand->action, ol !! 1193 rcu_assign_pointer(tsk->sighand, newsighand);
1182 sizeof(newsighand->act <<
1183 rcu_assign_pointer(me->sighan <<
1184 spin_unlock(&oldsighand->sigl 1194 spin_unlock(&oldsighand->siglock);
1185 write_unlock_irq(&tasklist_lo 1195 write_unlock_irq(&tasklist_lock);
1186 1196
1187 __cleanup_sighand(oldsighand) 1197 __cleanup_sighand(oldsighand);
1188 } 1198 }
>> 1199
>> 1200 BUG_ON(!thread_group_leader(tsk));
1189 return 0; 1201 return 0;
>> 1202
>> 1203 killed:
>> 1204 /* protects against exit_notify() and __exit_signal() */
>> 1205 read_lock(&tasklist_lock);
>> 1206 sig->group_exit_task = NULL;
>> 1207 sig->notify_count = 0;
>> 1208 read_unlock(&tasklist_lock);
>> 1209 return -EAGAIN;
1190 } 1210 }
1191 1211
1192 char *__get_task_comm(char *buf, size_t buf_s !! 1212 char *get_task_comm(char *buf, struct task_struct *tsk)
1193 { 1213 {
>> 1214 /* buf must be at least sizeof(tsk->comm) in size */
1194 task_lock(tsk); 1215 task_lock(tsk);
1195 /* Always NUL terminated and zero-pad !! 1216 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1196 strscpy_pad(buf, tsk->comm, buf_size) <<
1197 task_unlock(tsk); 1217 task_unlock(tsk);
1198 return buf; 1218 return buf;
1199 } 1219 }
1200 EXPORT_SYMBOL_GPL(__get_task_comm); !! 1220 EXPORT_SYMBOL_GPL(get_task_comm);
1201 1221
1202 /* 1222 /*
1203 * These functions flushes out all traces of 1223 * These functions flushes out all traces of the currently running executable
1204 * so that a new one can be started 1224 * so that a new one can be started
1205 */ 1225 */
1206 1226
1207 void __set_task_comm(struct task_struct *tsk, 1227 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1208 { 1228 {
1209 task_lock(tsk); 1229 task_lock(tsk);
1210 trace_task_rename(tsk, buf); 1230 trace_task_rename(tsk, buf);
1211 strscpy_pad(tsk->comm, buf, sizeof(ts !! 1231 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1212 task_unlock(tsk); 1232 task_unlock(tsk);
1213 perf_event_comm(tsk, exec); 1233 perf_event_comm(tsk, exec);
1214 } 1234 }
1215 1235
1216 /* !! 1236 int flush_old_exec(struct linux_binprm * bprm)
1217 * Calling this is the point of no return. No <<
1218 * seen by userspace since either the process <<
1219 * signal (via de_thread() or coredump), or w <<
1220 * (after exec_mmap()) by search_binary_handl <<
1221 */ <<
1222 int begin_new_exec(struct linux_binprm * bprm <<
1223 { 1237 {
1224 struct task_struct *me = current; <<
1225 int retval; 1238 int retval;
1226 1239
1227 /* Once we are committed compute the <<
1228 retval = bprm_creds_from_file(bprm); <<
1229 if (retval) <<
1230 return retval; <<
1231 <<
1232 /* <<
1233 * This tracepoint marks the point be <<
1234 * the current task is still unchange <<
1235 * no return). The later "sched_proce <<
1236 * the current task has successfully <<
1237 */ <<
1238 trace_sched_prepare_exec(current, bpr <<
1239 <<
1240 /* 1240 /*
1241 * Ensure all future errors are fatal !! 1241 * Make sure we have a private signal table and that
>> 1242 * we are unassociated from the previous thread group.
1242 */ 1243 */
1243 bprm->point_of_no_return = true; !! 1244 retval = de_thread(current);
1244 <<
1245 /* <<
1246 * Make this the only thread in the t <<
1247 */ <<
1248 retval = de_thread(me); <<
1249 if (retval) <<
1250 goto out; <<
1251 <<
1252 /* <<
1253 * Cancel any io_uring activity acros <<
1254 */ <<
1255 io_uring_task_cancel(); <<
1256 <<
1257 /* Ensure the files table is not shar <<
1258 retval = unshare_files(); <<
1259 if (retval) 1245 if (retval)
1260 goto out; 1246 goto out;
1261 1247
1262 /* 1248 /*
1263 * Must be called _before_ exec_mmap( 1249 * Must be called _before_ exec_mmap() as bprm->mm is
1264 * not visible until then. Doing it h !! 1250 * not visibile until then. This also enables the update
1265 * we don't race against replace_mm_e !! 1251 * to be lockless.
1266 */ 1252 */
1267 retval = set_mm_exe_file(bprm->mm, bp !! 1253 set_mm_exe_file(bprm->mm, bprm->file);
1268 if (retval) <<
1269 goto out; <<
1270 <<
1271 /* If the binary is not readable then <<
1272 would_dump(bprm, bprm->file); <<
1273 if (bprm->have_execfd) <<
1274 would_dump(bprm, bprm->execut <<
1275 1254
1276 /* 1255 /*
1277 * Release all of the old mmap stuff 1256 * Release all of the old mmap stuff
1278 */ 1257 */
1279 acct_arg_size(bprm, 0); 1258 acct_arg_size(bprm, 0);
1280 retval = exec_mmap(bprm->mm); 1259 retval = exec_mmap(bprm->mm);
1281 if (retval) 1260 if (retval)
1282 goto out; 1261 goto out;
1283 1262
1284 bprm->mm = NULL; !! 1263 bprm->mm = NULL; /* We're using it now */
1285 <<
1286 retval = exec_task_namespaces(); <<
1287 if (retval) <<
1288 goto out_unlock; <<
1289 <<
1290 #ifdef CONFIG_POSIX_TIMERS <<
1291 spin_lock_irq(&me->sighand->siglock); <<
1292 posix_cpu_timers_exit(me); <<
1293 spin_unlock_irq(&me->sighand->siglock <<
1294 exit_itimers(me); <<
1295 flush_itimer_signals(); <<
1296 #endif <<
1297 <<
1298 /* <<
1299 * Make the signal table private. <<
1300 */ <<
1301 retval = unshare_sighand(me); <<
1302 if (retval) <<
1303 goto out_unlock; <<
1304 1264
1305 me->flags &= ~(PF_RANDOMIZE | PF_FORK !! 1265 set_fs(USER_DS);
>> 1266 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1306 PF_NO 1267 PF_NOFREEZE | PF_NO_SETAFFINITY);
1307 flush_thread(); 1268 flush_thread();
1308 me->personality &= ~bprm->per_clear; !! 1269 current->personality &= ~bprm->per_clear;
1309 <<
1310 clear_syscall_work_syscall_user_dispa <<
1311 1270
1312 /* 1271 /*
1313 * We have to apply CLOEXEC before we 1272 * We have to apply CLOEXEC before we change whether the process is
1314 * dumpable (in setup_new_exec) to av 1273 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1315 * trying to access the should-be-clo 1274 * trying to access the should-be-closed file descriptors of a process
1316 * undergoing exec(2). 1275 * undergoing exec(2).
1317 */ 1276 */
1318 do_close_on_exec(me->files); !! 1277 do_close_on_exec(current->files);
1319 <<
1320 if (bprm->secureexec) { <<
1321 /* Make sure parent cannot si <<
1322 me->pdeath_signal = 0; <<
1323 <<
1324 /* <<
1325 * For secureexec, reset the <<
1326 * avoid bad behavior from th <<
1327 * happen before arch_pick_mm <<
1328 * RLIMIT_STACK, but after th <<
1329 * needing to clean up the ch <<
1330 */ <<
1331 if (bprm->rlim_stack.rlim_cur <<
1332 bprm->rlim_stack.rlim <<
1333 } <<
1334 <<
1335 me->sas_ss_sp = me->sas_ss_size = 0; <<
1336 <<
1337 /* <<
1338 * Figure out dumpability. Note that <<
1339 * is wrong, but userspace depends on <<
1340 * bprm->secureexec instead. <<
1341 */ <<
1342 if (bprm->interp_flags & BINPRM_FLAGS <<
1343 !(uid_eq(current_euid(), current_ <<
1344 gid_eq(current_egid(), current_ <<
1345 set_dumpable(current->mm, sui <<
1346 else <<
1347 set_dumpable(current->mm, SUI <<
1348 <<
1349 perf_event_exec(); <<
1350 __set_task_comm(me, kbasename(bprm->f <<
1351 <<
1352 /* An exec changes our domain. We are <<
1353 group */ <<
1354 WRITE_ONCE(me->self_exec_id, me->self <<
1355 flush_signal_handlers(me, 0); <<
1356 <<
1357 retval = set_cred_ucounts(bprm->cred) <<
1358 if (retval < 0) <<
1359 goto out_unlock; <<
1360 <<
1361 /* <<
1362 * install the new credentials for th <<
1363 */ <<
1364 security_bprm_committing_creds(bprm); <<
1365 <<
1366 commit_creds(bprm->cred); <<
1367 bprm->cred = NULL; <<
1368 <<
1369 /* <<
1370 * Disable monitoring for regular use <<
1371 * when executing setuid binaries. Mu <<
1372 * wait until new credentials are com <<
1373 * by commit_creds() above <<
1374 */ <<
1375 if (get_dumpable(me->mm) != SUID_DUMP <<
1376 perf_event_exit_task(me); <<
1377 /* <<
1378 * cred_guard_mutex must be held at l <<
1379 * ptrace_attach() from altering our <<
1380 * credentials; any time after this i <<
1381 */ <<
1382 security_bprm_committed_creds(bprm); <<
1383 <<
1384 /* Pass the opened binary to the inte <<
1385 if (bprm->have_execfd) { <<
1386 retval = get_unused_fd_flags( <<
1387 if (retval < 0) <<
1388 goto out_unlock; <<
1389 fd_install(retval, bprm->exec <<
1390 bprm->executable = NULL; <<
1391 bprm->execfd = retval; <<
1392 } <<
1393 return 0; 1278 return 0;
1394 1279
1395 out_unlock: <<
1396 up_write(&me->signal->exec_update_loc <<
1397 if (!bprm->cred) <<
1398 mutex_unlock(&me->signal->cre <<
1399 <<
1400 out: 1280 out:
1401 return retval; 1281 return retval;
1402 } 1282 }
1403 EXPORT_SYMBOL(begin_new_exec); !! 1283 EXPORT_SYMBOL(flush_old_exec);
1404 1284
1405 void would_dump(struct linux_binprm *bprm, st 1285 void would_dump(struct linux_binprm *bprm, struct file *file)
1406 { 1286 {
1407 struct inode *inode = file_inode(file 1287 struct inode *inode = file_inode(file);
1408 struct mnt_idmap *idmap = file_mnt_id !! 1288 if (inode_permission(inode, MAY_READ) < 0) {
1409 if (inode_permission(idmap, inode, MA <<
1410 struct user_namespace *old, * 1289 struct user_namespace *old, *user_ns;
1411 bprm->interp_flags |= BINPRM_ 1290 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1412 1291
1413 /* Ensure mm->user_ns contain 1292 /* Ensure mm->user_ns contains the executable */
1414 user_ns = old = bprm->mm->use 1293 user_ns = old = bprm->mm->user_ns;
1415 while ((user_ns != &init_user 1294 while ((user_ns != &init_user_ns) &&
1416 !privileged_wrt_inode_ !! 1295 !privileged_wrt_inode_uidgid(user_ns, inode))
1417 user_ns = user_ns->pa 1296 user_ns = user_ns->parent;
1418 1297
1419 if (old != user_ns) { 1298 if (old != user_ns) {
1420 bprm->mm->user_ns = g 1299 bprm->mm->user_ns = get_user_ns(user_ns);
1421 put_user_ns(old); 1300 put_user_ns(old);
1422 } 1301 }
1423 } 1302 }
1424 } 1303 }
1425 EXPORT_SYMBOL(would_dump); 1304 EXPORT_SYMBOL(would_dump);
1426 1305
1427 void setup_new_exec(struct linux_binprm * bpr 1306 void setup_new_exec(struct linux_binprm * bprm)
1428 { 1307 {
1429 /* Setup things that can depend upon !! 1308 arch_pick_mmap_layout(current->mm);
1430 struct task_struct *me = current; <<
1431 1309
1432 arch_pick_mmap_layout(me->mm, &bprm-> !! 1310 /* This is the point of no return */
>> 1311 current->sas_ss_sp = current->sas_ss_size = 0;
1433 1312
1434 arch_setup_new_exec(); !! 1313 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
>> 1314 set_dumpable(current->mm, SUID_DUMP_USER);
>> 1315 else
>> 1316 set_dumpable(current->mm, suid_dumpable);
>> 1317
>> 1318 perf_event_exec();
>> 1319 __set_task_comm(current, kbasename(bprm->filename), true);
1435 1320
1436 /* Set the new mm task size. We have 1321 /* Set the new mm task size. We have to do that late because it may
1437 * depend on TIF_32BIT which is only 1322 * depend on TIF_32BIT which is only updated in flush_thread() on
1438 * some architectures like powerpc 1323 * some architectures like powerpc
1439 */ 1324 */
1440 me->mm->task_size = TASK_SIZE; !! 1325 current->mm->task_size = TASK_SIZE;
1441 up_write(&me->signal->exec_update_loc <<
1442 mutex_unlock(&me->signal->cred_guard_ <<
1443 } <<
1444 EXPORT_SYMBOL(setup_new_exec); <<
1445 1326
1446 /* Runs immediately before start_thread() tak !! 1327 /* install the new credentials */
1447 void finalize_exec(struct linux_binprm *bprm) !! 1328 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1448 { !! 1329 !gid_eq(bprm->cred->gid, current_egid())) {
1449 /* Store any stack rlimit changes bef !! 1330 current->pdeath_signal = 0;
1450 task_lock(current->group_leader); !! 1331 } else {
1451 current->signal->rlim[RLIMIT_STACK] = !! 1332 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1452 task_unlock(current->group_leader); !! 1333 set_dumpable(current->mm, suid_dumpable);
>> 1334 }
>> 1335
>> 1336 /* An exec changes our domain. We are no longer part of the thread
>> 1337 group */
>> 1338 current->self_exec_id++;
>> 1339 flush_signal_handlers(current, 0);
1453 } 1340 }
1454 EXPORT_SYMBOL(finalize_exec); !! 1341 EXPORT_SYMBOL(setup_new_exec);
1455 1342
1456 /* 1343 /*
1457 * Prepare credentials and lock ->cred_guard_ 1344 * Prepare credentials and lock ->cred_guard_mutex.
1458 * setup_new_exec() commits the new creds and !! 1345 * install_exec_creds() commits the new creds and drops the lock.
1459 * Or, if exec fails before, free_bprm() shou !! 1346 * Or, if exec fails before, free_bprm() should release ->cred and
1460 * and unlock. 1347 * and unlock.
1461 */ 1348 */
1462 static int prepare_bprm_creds(struct linux_bi !! 1349 int prepare_bprm_creds(struct linux_binprm *bprm)
1463 { 1350 {
1464 if (mutex_lock_interruptible(¤t 1351 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1465 return -ERESTARTNOINTR; 1352 return -ERESTARTNOINTR;
1466 1353
1467 bprm->cred = prepare_exec_creds(); 1354 bprm->cred = prepare_exec_creds();
1468 if (likely(bprm->cred)) 1355 if (likely(bprm->cred))
1469 return 0; 1356 return 0;
1470 1357
1471 mutex_unlock(¤t->signal->cred_g 1358 mutex_unlock(¤t->signal->cred_guard_mutex);
1472 return -ENOMEM; 1359 return -ENOMEM;
1473 } 1360 }
1474 1361
1475 /* Matches do_open_execat() */ <<
1476 static void do_close_execat(struct file *file <<
1477 { <<
1478 if (file) <<
1479 fput(file); <<
1480 } <<
1481 <<
1482 static void free_bprm(struct linux_binprm *bp 1362 static void free_bprm(struct linux_binprm *bprm)
1483 { 1363 {
1484 if (bprm->mm) { <<
1485 acct_arg_size(bprm, 0); <<
1486 mmput(bprm->mm); <<
1487 } <<
1488 free_arg_pages(bprm); 1364 free_arg_pages(bprm);
1489 if (bprm->cred) { 1365 if (bprm->cred) {
1490 mutex_unlock(¤t->signal 1366 mutex_unlock(¤t->signal->cred_guard_mutex);
1491 abort_creds(bprm->cred); 1367 abort_creds(bprm->cred);
1492 } 1368 }
1493 do_close_execat(bprm->file); !! 1369 if (bprm->file) {
1494 if (bprm->executable) !! 1370 allow_write_access(bprm->file);
1495 fput(bprm->executable); !! 1371 fput(bprm->file);
>> 1372 }
1496 /* If a binfmt changed the interp, fr 1373 /* If a binfmt changed the interp, free it. */
1497 if (bprm->interp != bprm->filename) 1374 if (bprm->interp != bprm->filename)
1498 kfree(bprm->interp); 1375 kfree(bprm->interp);
1499 kfree(bprm->fdpath); <<
1500 kfree(bprm); 1376 kfree(bprm);
1501 } 1377 }
1502 1378
1503 static struct linux_binprm *alloc_bprm(int fd !! 1379 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1504 { <<
1505 struct linux_binprm *bprm; <<
1506 struct file *file; <<
1507 int retval = -ENOMEM; <<
1508 <<
1509 file = do_open_execat(fd, filename, f <<
1510 if (IS_ERR(file)) <<
1511 return ERR_CAST(file); <<
1512 <<
1513 bprm = kzalloc(sizeof(*bprm), GFP_KER <<
1514 if (!bprm) { <<
1515 do_close_execat(file); <<
1516 return ERR_PTR(-ENOMEM); <<
1517 } <<
1518 <<
1519 bprm->file = file; <<
1520 <<
1521 if (fd == AT_FDCWD || filename->name[ <<
1522 bprm->filename = filename->na <<
1523 } else { <<
1524 if (filename->name[0] == '\0' <<
1525 bprm->fdpath = kaspri <<
1526 else <<
1527 bprm->fdpath = kaspri <<
1528 <<
1529 if (!bprm->fdpath) <<
1530 goto out_free; <<
1531 <<
1532 /* <<
1533 * Record that a name derived <<
1534 * inaccessible after exec. <<
1535 * choose to fail when the ex <<
1536 * interpreter and an open fi <<
1537 * the interpreter. This mak <<
1538 * than having the interprete <<
1539 * when it finds the executab <<
1540 */ <<
1541 if (get_close_on_exec(fd)) <<
1542 bprm->interp_flags |= <<
1543 <<
1544 bprm->filename = bprm->fdpath <<
1545 } <<
1546 bprm->interp = bprm->filename; <<
1547 <<
1548 retval = bprm_mm_init(bprm); <<
1549 if (!retval) <<
1550 return bprm; <<
1551 <<
1552 out_free: <<
1553 free_bprm(bprm); <<
1554 return ERR_PTR(retval); <<
1555 } <<
1556 <<
1557 int bprm_change_interp(const char *interp, st <<
1558 { 1380 {
1559 /* If a binfmt changed the interp, fr 1381 /* If a binfmt changed the interp, free it first. */
1560 if (bprm->interp != bprm->filename) 1382 if (bprm->interp != bprm->filename)
1561 kfree(bprm->interp); 1383 kfree(bprm->interp);
1562 bprm->interp = kstrdup(interp, GFP_KE 1384 bprm->interp = kstrdup(interp, GFP_KERNEL);
1563 if (!bprm->interp) 1385 if (!bprm->interp)
1564 return -ENOMEM; 1386 return -ENOMEM;
1565 return 0; 1387 return 0;
1566 } 1388 }
1567 EXPORT_SYMBOL(bprm_change_interp); 1389 EXPORT_SYMBOL(bprm_change_interp);
1568 1390
1569 /* 1391 /*
>> 1392 * install the new credentials for this executable
>> 1393 */
>> 1394 void install_exec_creds(struct linux_binprm *bprm)
>> 1395 {
>> 1396 security_bprm_committing_creds(bprm);
>> 1397
>> 1398 commit_creds(bprm->cred);
>> 1399 bprm->cred = NULL;
>> 1400
>> 1401 /*
>> 1402 * Disable monitoring for regular users
>> 1403 * when executing setuid binaries. Must
>> 1404 * wait until new credentials are committed
>> 1405 * by commit_creds() above
>> 1406 */
>> 1407 if (get_dumpable(current->mm) != SUID_DUMP_USER)
>> 1408 perf_event_exit_task(current);
>> 1409 /*
>> 1410 * cred_guard_mutex must be held at least to this point to prevent
>> 1411 * ptrace_attach() from altering our determination of the task's
>> 1412 * credentials; any time after this it may be unlocked.
>> 1413 */
>> 1414 security_bprm_committed_creds(bprm);
>> 1415 mutex_unlock(¤t->signal->cred_guard_mutex);
>> 1416 }
>> 1417 EXPORT_SYMBOL(install_exec_creds);
>> 1418
>> 1419 /*
1570 * determine how safe it is to execute the pr 1420 * determine how safe it is to execute the proposed program
1571 * - the caller must hold ->cred_guard_mutex 1421 * - the caller must hold ->cred_guard_mutex to protect against
1572 * PTRACE_ATTACH or seccomp thread-sync 1422 * PTRACE_ATTACH or seccomp thread-sync
1573 */ 1423 */
1574 static void check_unsafe_exec(struct linux_bi 1424 static void check_unsafe_exec(struct linux_binprm *bprm)
1575 { 1425 {
1576 struct task_struct *p = current, *t; 1426 struct task_struct *p = current, *t;
1577 unsigned n_fs; 1427 unsigned n_fs;
1578 1428
1579 if (p->ptrace) !! 1429 if (p->ptrace) {
1580 bprm->unsafe |= LSM_UNSAFE_PT !! 1430 if (ptracer_capable(p, current_user_ns()))
>> 1431 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
>> 1432 else
>> 1433 bprm->unsafe |= LSM_UNSAFE_PTRACE;
>> 1434 }
1581 1435
1582 /* 1436 /*
1583 * This isn't strictly necessary, but 1437 * This isn't strictly necessary, but it makes it harder for LSMs to
1584 * mess up. 1438 * mess up.
1585 */ 1439 */
1586 if (task_no_new_privs(current)) 1440 if (task_no_new_privs(current))
1587 bprm->unsafe |= LSM_UNSAFE_NO 1441 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1588 1442
1589 /* !! 1443 t = p;
1590 * If another task is sharing our fs, <<
1591 * suid exec because the differently <<
1592 * will be able to manipulate the cur <<
1593 * It would be nice to force an unsha <<
1594 */ <<
1595 n_fs = 1; 1444 n_fs = 1;
1596 spin_lock(&p->fs->lock); 1445 spin_lock(&p->fs->lock);
1597 rcu_read_lock(); 1446 rcu_read_lock();
1598 for_other_threads(p, t) { !! 1447 while_each_thread(p, t) {
1599 if (t->fs == p->fs) 1448 if (t->fs == p->fs)
1600 n_fs++; 1449 n_fs++;
1601 } 1450 }
1602 rcu_read_unlock(); 1451 rcu_read_unlock();
1603 1452
1604 /* "users" and "in_exec" locked for c <<
1605 if (p->fs->users > n_fs) 1453 if (p->fs->users > n_fs)
1606 bprm->unsafe |= LSM_UNSAFE_SH 1454 bprm->unsafe |= LSM_UNSAFE_SHARE;
1607 else 1455 else
1608 p->fs->in_exec = 1; 1456 p->fs->in_exec = 1;
1609 spin_unlock(&p->fs->lock); 1457 spin_unlock(&p->fs->lock);
1610 } 1458 }
1611 1459
1612 static void bprm_fill_uid(struct linux_binprm !! 1460 static void bprm_fill_uid(struct linux_binprm *bprm)
1613 { 1461 {
1614 /* Handle suid and sgid on files */ !! 1462 struct inode *inode;
1615 struct mnt_idmap *idmap; <<
1616 struct inode *inode = file_inode(file <<
1617 unsigned int mode; 1463 unsigned int mode;
1618 vfsuid_t vfsuid; !! 1464 kuid_t uid;
1619 vfsgid_t vfsgid; !! 1465 kgid_t gid;
1620 int err; <<
1621 1466
1622 if (!mnt_may_suid(file->f_path.mnt)) !! 1467 /*
>> 1468 * Since this can be called multiple times (via prepare_binprm),
>> 1469 * we must clear any previous work done when setting set[ug]id
>> 1470 * bits from any earlier bprm->file uses (for example when run
>> 1471 * first for a setuid script then again for its interpreter).
>> 1472 */
>> 1473 bprm->cred->euid = current_euid();
>> 1474 bprm->cred->egid = current_egid();
>> 1475
>> 1476 if (!mnt_may_suid(bprm->file->f_path.mnt))
1623 return; 1477 return;
1624 1478
1625 if (task_no_new_privs(current)) 1479 if (task_no_new_privs(current))
1626 return; 1480 return;
1627 1481
>> 1482 inode = file_inode(bprm->file);
1628 mode = READ_ONCE(inode->i_mode); 1483 mode = READ_ONCE(inode->i_mode);
1629 if (!(mode & (S_ISUID|S_ISGID))) 1484 if (!(mode & (S_ISUID|S_ISGID)))
1630 return; 1485 return;
1631 1486
1632 idmap = file_mnt_idmap(file); <<
1633 <<
1634 /* Be careful if suid/sgid is set */ 1487 /* Be careful if suid/sgid is set */
1635 inode_lock(inode); 1488 inode_lock(inode);
1636 1489
1637 /* Atomically reload and check mode/u !! 1490 /* reload atomically mode/uid/gid now that lock held */
1638 mode = inode->i_mode; 1491 mode = inode->i_mode;
1639 vfsuid = i_uid_into_vfsuid(idmap, ino !! 1492 uid = inode->i_uid;
1640 vfsgid = i_gid_into_vfsgid(idmap, ino !! 1493 gid = inode->i_gid;
1641 err = inode_permission(idmap, inode, <<
1642 inode_unlock(inode); 1494 inode_unlock(inode);
1643 1495
1644 /* Did the exec bit vanish out from u <<
1645 if (err) <<
1646 return; <<
1647 <<
1648 /* We ignore suid/sgid if there are n 1496 /* We ignore suid/sgid if there are no mappings for them in the ns */
1649 if (!vfsuid_has_mapping(bprm->cred->u !! 1497 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1650 !vfsgid_has_mapping(bprm->cred->u !! 1498 !kgid_has_mapping(bprm->cred->user_ns, gid))
1651 return; 1499 return;
1652 1500
1653 if (mode & S_ISUID) { 1501 if (mode & S_ISUID) {
1654 bprm->per_clear |= PER_CLEAR_ 1502 bprm->per_clear |= PER_CLEAR_ON_SETID;
1655 bprm->cred->euid = vfsuid_int !! 1503 bprm->cred->euid = uid;
1656 } 1504 }
1657 1505
1658 if ((mode & (S_ISGID | S_IXGRP)) == ( 1506 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1659 bprm->per_clear |= PER_CLEAR_ 1507 bprm->per_clear |= PER_CLEAR_ON_SETID;
1660 bprm->cred->egid = vfsgid_int !! 1508 bprm->cred->egid = gid;
1661 } 1509 }
1662 } 1510 }
1663 1511
1664 /* 1512 /*
1665 * Compute brpm->cred based upon the final bi <<
1666 */ <<
1667 static int bprm_creds_from_file(struct linux_ <<
1668 { <<
1669 /* Compute creds based on which file? <<
1670 struct file *file = bprm->execfd_cred <<
1671 <<
1672 bprm_fill_uid(bprm, file); <<
1673 return security_bprm_creds_from_file( <<
1674 } <<
1675 <<
1676 /* <<
1677 * Fill the binprm structure from the inode. 1513 * Fill the binprm structure from the inode.
1678 * Read the first BINPRM_BUF_SIZE bytes !! 1514 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1679 * 1515 *
1680 * This may be called multiple times for bina 1516 * This may be called multiple times for binary chains (scripts for example).
1681 */ 1517 */
1682 static int prepare_binprm(struct linux_binprm !! 1518 int prepare_binprm(struct linux_binprm *bprm)
1683 { 1519 {
1684 loff_t pos = 0; !! 1520 int retval;
>> 1521
>> 1522 bprm_fill_uid(bprm);
>> 1523
>> 1524 /* fill in binprm security blob */
>> 1525 retval = security_bprm_set_creds(bprm);
>> 1526 if (retval)
>> 1527 return retval;
>> 1528 bprm->cred_prepared = 1;
1685 1529
1686 memset(bprm->buf, 0, BINPRM_BUF_SIZE) 1530 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1687 return kernel_read(bprm->file, bprm-> !! 1531 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1688 } 1532 }
1689 1533
>> 1534 EXPORT_SYMBOL(prepare_binprm);
>> 1535
1690 /* 1536 /*
1691 * Arguments are '\0' separated strings found 1537 * Arguments are '\0' separated strings found at the location bprm->p
1692 * points to; chop off the first by relocatin 1538 * points to; chop off the first by relocating brpm->p to right after
1693 * the first '\0' encountered. 1539 * the first '\0' encountered.
1694 */ 1540 */
1695 int remove_arg_zero(struct linux_binprm *bprm 1541 int remove_arg_zero(struct linux_binprm *bprm)
1696 { 1542 {
>> 1543 int ret = 0;
1697 unsigned long offset; 1544 unsigned long offset;
1698 char *kaddr; 1545 char *kaddr;
1699 struct page *page; 1546 struct page *page;
1700 1547
1701 if (!bprm->argc) 1548 if (!bprm->argc)
1702 return 0; 1549 return 0;
1703 1550
1704 do { 1551 do {
1705 offset = bprm->p & ~PAGE_MASK 1552 offset = bprm->p & ~PAGE_MASK;
1706 page = get_arg_page(bprm, bpr 1553 page = get_arg_page(bprm, bprm->p, 0);
1707 if (!page) !! 1554 if (!page) {
1708 return -EFAULT; !! 1555 ret = -EFAULT;
1709 kaddr = kmap_local_page(page) !! 1556 goto out;
>> 1557 }
>> 1558 kaddr = kmap_atomic(page);
1710 1559
1711 for (; offset < PAGE_SIZE && 1560 for (; offset < PAGE_SIZE && kaddr[offset];
1712 offset++, bpr 1561 offset++, bprm->p++)
1713 ; 1562 ;
1714 1563
1715 kunmap_local(kaddr); !! 1564 kunmap_atomic(kaddr);
1716 put_arg_page(page); 1565 put_arg_page(page);
1717 } while (offset == PAGE_SIZE); 1566 } while (offset == PAGE_SIZE);
1718 1567
1719 bprm->p++; 1568 bprm->p++;
1720 bprm->argc--; 1569 bprm->argc--;
>> 1570 ret = 0;
1721 1571
1722 return 0; !! 1572 out:
>> 1573 return ret;
1723 } 1574 }
1724 EXPORT_SYMBOL(remove_arg_zero); 1575 EXPORT_SYMBOL(remove_arg_zero);
1725 1576
1726 #define printable(c) (((c)=='\t') || ((c)=='\ 1577 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1727 /* 1578 /*
1728 * cycle the list of binary formats handler, 1579 * cycle the list of binary formats handler, until one recognizes the image
1729 */ 1580 */
1730 static int search_binary_handler(struct linux !! 1581 int search_binary_handler(struct linux_binprm *bprm)
1731 { 1582 {
1732 bool need_retry = IS_ENABLED(CONFIG_M 1583 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1733 struct linux_binfmt *fmt; 1584 struct linux_binfmt *fmt;
1734 int retval; 1585 int retval;
1735 1586
1736 retval = prepare_binprm(bprm); !! 1587 /* This allows 4 levels of binfmt rewrites before failing hard. */
1737 if (retval < 0) !! 1588 if (bprm->recursion_depth > 5)
1738 return retval; !! 1589 return -ELOOP;
1739 1590
1740 retval = security_bprm_check(bprm); 1591 retval = security_bprm_check(bprm);
1741 if (retval) 1592 if (retval)
1742 return retval; 1593 return retval;
1743 1594
1744 retval = -ENOENT; 1595 retval = -ENOENT;
1745 retry: 1596 retry:
1746 read_lock(&binfmt_lock); 1597 read_lock(&binfmt_lock);
1747 list_for_each_entry(fmt, &formats, lh 1598 list_for_each_entry(fmt, &formats, lh) {
1748 if (!try_module_get(fmt->modu 1599 if (!try_module_get(fmt->module))
1749 continue; 1600 continue;
1750 read_unlock(&binfmt_lock); 1601 read_unlock(&binfmt_lock);
1751 !! 1602 bprm->recursion_depth++;
1752 retval = fmt->load_binary(bpr 1603 retval = fmt->load_binary(bprm);
1753 <<
1754 read_lock(&binfmt_lock); 1604 read_lock(&binfmt_lock);
1755 put_binfmt(fmt); 1605 put_binfmt(fmt);
1756 if (bprm->point_of_no_return !! 1606 bprm->recursion_depth--;
>> 1607 if (retval < 0 && !bprm->mm) {
>> 1608 /* we got to flush_old_exec() and failed after it */
>> 1609 read_unlock(&binfmt_lock);
>> 1610 force_sigsegv(SIGSEGV, current);
>> 1611 return retval;
>> 1612 }
>> 1613 if (retval != -ENOEXEC || !bprm->file) {
1757 read_unlock(&binfmt_l 1614 read_unlock(&binfmt_lock);
1758 return retval; 1615 return retval;
1759 } 1616 }
1760 } 1617 }
1761 read_unlock(&binfmt_lock); 1618 read_unlock(&binfmt_lock);
1762 1619
1763 if (need_retry) { 1620 if (need_retry) {
1764 if (printable(bprm->buf[0]) & 1621 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1765 printable(bprm->buf[2]) & 1622 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1766 return retval; 1623 return retval;
1767 if (request_module("binfmt-%0 1624 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1768 return retval; 1625 return retval;
1769 need_retry = false; 1626 need_retry = false;
1770 goto retry; 1627 goto retry;
1771 } 1628 }
1772 1629
1773 return retval; 1630 return retval;
1774 } 1631 }
>> 1632 EXPORT_SYMBOL(search_binary_handler);
1775 1633
1776 /* binfmt handlers will call back into begin_ <<
1777 static int exec_binprm(struct linux_binprm *b 1634 static int exec_binprm(struct linux_binprm *bprm)
1778 { 1635 {
1779 pid_t old_pid, old_vpid; 1636 pid_t old_pid, old_vpid;
1780 int ret, depth; !! 1637 int ret;
1781 1638
1782 /* Need to fetch pid before load_bina 1639 /* Need to fetch pid before load_binary changes it */
1783 old_pid = current->pid; 1640 old_pid = current->pid;
1784 rcu_read_lock(); 1641 rcu_read_lock();
1785 old_vpid = task_pid_nr_ns(current, ta 1642 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1786 rcu_read_unlock(); 1643 rcu_read_unlock();
1787 1644
1788 /* This allows 4 levels of binfmt rew !! 1645 ret = ccs_search_binary_handler(bprm);
1789 for (depth = 0;; depth++) { !! 1646 if (ret >= 0) {
1790 struct file *exec; !! 1647 audit_bprm(bprm);
1791 if (depth > 5) !! 1648 trace_sched_process_exec(current, old_pid, bprm);
1792 return -ELOOP; !! 1649 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1793 !! 1650 proc_exec_connector(current);
1794 ret = search_binary_handler(b <<
1795 if (ret < 0) <<
1796 return ret; <<
1797 if (!bprm->interpreter) <<
1798 break; <<
1799 <<
1800 exec = bprm->file; <<
1801 bprm->file = bprm->interprete <<
1802 bprm->interpreter = NULL; <<
1803 <<
1804 if (unlikely(bprm->have_execf <<
1805 if (bprm->executable) <<
1806 fput(exec); <<
1807 return -ENOEX <<
1808 } <<
1809 bprm->executable = ex <<
1810 } else <<
1811 fput(exec); <<
1812 } 1651 }
1813 1652
1814 audit_bprm(bprm); !! 1653 return ret;
1815 trace_sched_process_exec(current, old <<
1816 ptrace_event(PTRACE_EVENT_EXEC, old_v <<
1817 proc_exec_connector(current); <<
1818 return 0; <<
1819 } <<
1820 <<
1821 static int bprm_execve(struct linux_binprm *b <<
1822 { <<
1823 int retval; <<
1824 <<
1825 retval = prepare_bprm_creds(bprm); <<
1826 if (retval) <<
1827 return retval; <<
1828 <<
1829 /* <<
1830 * Check for unsafe execution states <<
1831 * will call back into begin_new_exec <<
1832 * where setuid-ness is evaluated. <<
1833 */ <<
1834 check_unsafe_exec(bprm); <<
1835 current->in_execve = 1; <<
1836 sched_mm_cid_before_execve(current); <<
1837 <<
1838 sched_exec(); <<
1839 <<
1840 /* Set the unchanging part of bprm->c <<
1841 retval = security_bprm_creds_for_exec <<
1842 if (retval) <<
1843 goto out; <<
1844 <<
1845 retval = ccs_exec_binprm(bprm); <<
1846 if (retval < 0) <<
1847 goto out; <<
1848 <<
1849 sched_mm_cid_after_execve(current); <<
1850 /* execve succeeded */ <<
1851 current->fs->in_exec = 0; <<
1852 current->in_execve = 0; <<
1853 rseq_execve(current); <<
1854 user_events_execve(current); <<
1855 acct_update_integrals(current); <<
1856 task_numa_free(current, false); <<
1857 return retval; <<
1858 <<
1859 out: <<
1860 /* <<
1861 * If past the point of no return ens <<
1862 * returns to the userspace process. <<
1863 * signal if present otherwise termin <<
1864 * SIGSEGV. <<
1865 */ <<
1866 if (bprm->point_of_no_return && !fata <<
1867 force_fatal_sig(SIGSEGV); <<
1868 <<
1869 sched_mm_cid_after_execve(current); <<
1870 current->fs->in_exec = 0; <<
1871 current->in_execve = 0; <<
1872 <<
1873 return retval; <<
1874 } 1654 }
1875 1655
>> 1656 /*
>> 1657 * sys_execve() executes a new program.
>> 1658 */
1876 static int do_execveat_common(int fd, struct 1659 static int do_execveat_common(int fd, struct filename *filename,
1877 struct user_arg 1660 struct user_arg_ptr argv,
1878 struct user_arg 1661 struct user_arg_ptr envp,
1879 int flags) 1662 int flags)
1880 { 1663 {
>> 1664 char *pathbuf = NULL;
1881 struct linux_binprm *bprm; 1665 struct linux_binprm *bprm;
>> 1666 struct file *file;
>> 1667 struct files_struct *displaced;
1882 int retval; 1668 int retval;
1883 1669
1884 if (IS_ERR(filename)) 1670 if (IS_ERR(filename))
1885 return PTR_ERR(filename); 1671 return PTR_ERR(filename);
1886 1672
1887 /* 1673 /*
1888 * We move the actual failure in case 1674 * We move the actual failure in case of RLIMIT_NPROC excess from
1889 * set*uid() to execve() because too 1675 * set*uid() to execve() because too many poorly written programs
1890 * don't check setuid() return code. 1676 * don't check setuid() return code. Here we additionally recheck
1891 * whether NPROC limit is still excee 1677 * whether NPROC limit is still exceeded.
1892 */ 1678 */
1893 if ((current->flags & PF_NPROC_EXCEED 1679 if ((current->flags & PF_NPROC_EXCEEDED) &&
1894 is_rlimit_overlimit(current_ucoun !! 1680 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) {
1895 retval = -EAGAIN; 1681 retval = -EAGAIN;
1896 goto out_ret; 1682 goto out_ret;
1897 } 1683 }
1898 1684
1899 /* We're below the limit (still or ag 1685 /* We're below the limit (still or again), so we don't want to make
1900 * further execve() calls fail. */ 1686 * further execve() calls fail. */
1901 current->flags &= ~PF_NPROC_EXCEEDED; 1687 current->flags &= ~PF_NPROC_EXCEEDED;
1902 1688
1903 bprm = alloc_bprm(fd, filename, flags !! 1689 retval = unshare_files(&displaced);
1904 if (IS_ERR(bprm)) { !! 1690 if (retval)
1905 retval = PTR_ERR(bprm); <<
1906 goto out_ret; 1691 goto out_ret;
1907 } <<
1908 1692
1909 retval = count(argv, MAX_ARG_STRINGS) !! 1693 retval = -ENOMEM;
1910 if (retval == 0) !! 1694 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1911 pr_warn_once("process '%s' la !! 1695 if (!bprm)
1912 current->comm, b !! 1696 goto out_files;
1913 if (retval < 0) <<
1914 goto out_free; <<
1915 bprm->argc = retval; <<
1916 <<
1917 retval = count(envp, MAX_ARG_STRINGS) <<
1918 if (retval < 0) <<
1919 goto out_free; <<
1920 bprm->envc = retval; <<
1921 1697
1922 retval = bprm_stack_limits(bprm); !! 1698 retval = prepare_bprm_creds(bprm);
1923 if (retval < 0) !! 1699 if (retval)
1924 goto out_free; 1700 goto out_free;
1925 1701
1926 retval = copy_string_kernel(bprm->fil !! 1702 check_unsafe_exec(bprm);
1927 if (retval < 0) !! 1703 current->in_execve = 1;
1928 goto out_free; <<
1929 bprm->exec = bprm->p; <<
1930 1704
1931 retval = copy_strings(bprm->envc, env !! 1705 file = do_open_execat(fd, filename, flags);
1932 if (retval < 0) !! 1706 retval = PTR_ERR(file);
1933 goto out_free; !! 1707 if (IS_ERR(file))
>> 1708 goto out_unmark;
1934 1709
1935 retval = copy_strings(bprm->argc, arg !! 1710 sched_exec();
1936 if (retval < 0) <<
1937 goto out_free; <<
1938 1711
1939 /* !! 1712 bprm->file = file;
1940 * When argv is empty, add an empty s !! 1713 if (fd == AT_FDCWD || filename->name[0] == '/') {
1941 * ensure confused userspace programs !! 1714 bprm->filename = filename->name;
1942 * from argv[1] won't end up walking !! 1715 } else {
1943 * bprm_stack_limits(). !! 1716 if (filename->name[0] == '\0')
1944 */ !! 1717 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1945 if (bprm->argc == 0) { !! 1718 else
1946 retval = copy_string_kernel(" !! 1719 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1947 if (retval < 0) !! 1720 fd, filename->name);
1948 goto out_free; !! 1721 if (!pathbuf) {
1949 bprm->argc = 1; !! 1722 retval = -ENOMEM;
>> 1723 goto out_unmark;
>> 1724 }
>> 1725 /*
>> 1726 * Record that a name derived from an O_CLOEXEC fd will be
>> 1727 * inaccessible after exec. Relies on having exclusive access to
>> 1728 * current->files (due to unshare_files above).
>> 1729 */
>> 1730 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
>> 1731 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
>> 1732 bprm->filename = pathbuf;
1950 } 1733 }
>> 1734 bprm->interp = bprm->filename;
1951 1735
1952 retval = bprm_execve(bprm); !! 1736 retval = bprm_mm_init(bprm);
1953 out_free: !! 1737 if (retval)
1954 free_bprm(bprm); !! 1738 goto out_unmark;
1955 <<
1956 out_ret: <<
1957 putname(filename); <<
1958 return retval; <<
1959 } <<
1960 <<
1961 int kernel_execve(const char *kernel_filename <<
1962 const char *const *argv, co <<
1963 { <<
1964 struct filename *filename; <<
1965 struct linux_binprm *bprm; <<
1966 int fd = AT_FDCWD; <<
1967 int retval; <<
1968 <<
1969 /* It is non-sense for kernel threads <<
1970 if (WARN_ON_ONCE(current->flags & PF_ <<
1971 return -EINVAL; <<
1972 1739
1973 filename = getname_kernel(kernel_file !! 1740 bprm->argc = count(argv, MAX_ARG_STRINGS);
1974 if (IS_ERR(filename)) !! 1741 if ((retval = bprm->argc) < 0)
1975 return PTR_ERR(filename); !! 1742 goto out;
1976 1743
1977 bprm = alloc_bprm(fd, filename, 0); !! 1744 bprm->envc = count(envp, MAX_ARG_STRINGS);
1978 if (IS_ERR(bprm)) { !! 1745 if ((retval = bprm->envc) < 0)
1979 retval = PTR_ERR(bprm); !! 1746 goto out;
1980 goto out_ret; <<
1981 } <<
1982 1747
1983 retval = count_strings_kernel(argv); !! 1748 retval = prepare_binprm(bprm);
1984 if (WARN_ON_ONCE(retval == 0)) <<
1985 retval = -EINVAL; <<
1986 if (retval < 0) 1749 if (retval < 0)
1987 goto out_free; !! 1750 goto out;
1988 bprm->argc = retval; <<
1989 1751
1990 retval = count_strings_kernel(envp); !! 1752 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1991 if (retval < 0) 1753 if (retval < 0)
1992 goto out_free; !! 1754 goto out;
1993 bprm->envc = retval; <<
1994 1755
1995 retval = bprm_stack_limits(bprm); !! 1756 bprm->exec = bprm->p;
>> 1757 retval = copy_strings(bprm->envc, envp, bprm);
1996 if (retval < 0) 1758 if (retval < 0)
1997 goto out_free; !! 1759 goto out;
1998 1760
1999 retval = copy_string_kernel(bprm->fil !! 1761 retval = copy_strings(bprm->argc, argv, bprm);
2000 if (retval < 0) 1762 if (retval < 0)
2001 goto out_free; !! 1763 goto out;
2002 bprm->exec = bprm->p; <<
2003 1764
2004 retval = copy_strings_kernel(bprm->en !! 1765 would_dump(bprm, bprm->file);
2005 if (retval < 0) <<
2006 goto out_free; <<
2007 1766
2008 retval = copy_strings_kernel(bprm->ar !! 1767 retval = exec_binprm(bprm);
2009 if (retval < 0) 1768 if (retval < 0)
2010 goto out_free; !! 1769 goto out;
>> 1770
>> 1771 /* execve succeeded */
>> 1772 current->fs->in_exec = 0;
>> 1773 current->in_execve = 0;
>> 1774 acct_update_integrals(current);
>> 1775 task_numa_free(current);
>> 1776 free_bprm(bprm);
>> 1777 kfree(pathbuf);
>> 1778 putname(filename);
>> 1779 if (displaced)
>> 1780 put_files_struct(displaced);
>> 1781 return retval;
>> 1782
>> 1783 out:
>> 1784 if (bprm->mm) {
>> 1785 acct_arg_size(bprm, 0);
>> 1786 mmput(bprm->mm);
>> 1787 }
>> 1788
>> 1789 out_unmark:
>> 1790 current->fs->in_exec = 0;
>> 1791 current->in_execve = 0;
2011 1792
2012 retval = bprm_execve(bprm); <<
2013 out_free: 1793 out_free:
2014 free_bprm(bprm); 1794 free_bprm(bprm);
>> 1795 kfree(pathbuf);
>> 1796
>> 1797 out_files:
>> 1798 if (displaced)
>> 1799 reset_files_struct(displaced);
2015 out_ret: 1800 out_ret:
2016 putname(filename); 1801 putname(filename);
2017 return retval; 1802 return retval;
2018 } 1803 }
2019 1804
2020 static int do_execve(struct filename *filenam !! 1805 int do_execve(struct filename *filename,
2021 const char __user *const __user *__ar 1806 const char __user *const __user *__argv,
2022 const char __user *const __user *__en 1807 const char __user *const __user *__envp)
2023 { 1808 {
2024 struct user_arg_ptr argv = { .ptr.nat 1809 struct user_arg_ptr argv = { .ptr.native = __argv };
2025 struct user_arg_ptr envp = { .ptr.nat 1810 struct user_arg_ptr envp = { .ptr.native = __envp };
2026 return do_execveat_common(AT_FDCWD, f 1811 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2027 } 1812 }
2028 1813
2029 static int do_execveat(int fd, struct filenam !! 1814 int do_execveat(int fd, struct filename *filename,
2030 const char __user *const __us 1815 const char __user *const __user *__argv,
2031 const char __user *const __us 1816 const char __user *const __user *__envp,
2032 int flags) 1817 int flags)
2033 { 1818 {
2034 struct user_arg_ptr argv = { .ptr.nat 1819 struct user_arg_ptr argv = { .ptr.native = __argv };
2035 struct user_arg_ptr envp = { .ptr.nat 1820 struct user_arg_ptr envp = { .ptr.native = __envp };
2036 1821
2037 return do_execveat_common(fd, filenam 1822 return do_execveat_common(fd, filename, argv, envp, flags);
2038 } 1823 }
2039 1824
2040 #ifdef CONFIG_COMPAT 1825 #ifdef CONFIG_COMPAT
2041 static int compat_do_execve(struct filename * 1826 static int compat_do_execve(struct filename *filename,
2042 const compat_uptr_t __user *__argv, 1827 const compat_uptr_t __user *__argv,
2043 const compat_uptr_t __user *__envp) 1828 const compat_uptr_t __user *__envp)
2044 { 1829 {
2045 struct user_arg_ptr argv = { 1830 struct user_arg_ptr argv = {
2046 .is_compat = true, 1831 .is_compat = true,
2047 .ptr.compat = __argv, 1832 .ptr.compat = __argv,
2048 }; 1833 };
2049 struct user_arg_ptr envp = { 1834 struct user_arg_ptr envp = {
2050 .is_compat = true, 1835 .is_compat = true,
2051 .ptr.compat = __envp, 1836 .ptr.compat = __envp,
2052 }; 1837 };
2053 return do_execveat_common(AT_FDCWD, f 1838 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2054 } 1839 }
2055 1840
2056 static int compat_do_execveat(int fd, struct 1841 static int compat_do_execveat(int fd, struct filename *filename,
2057 const compat_up 1842 const compat_uptr_t __user *__argv,
2058 const compat_up 1843 const compat_uptr_t __user *__envp,
2059 int flags) 1844 int flags)
2060 { 1845 {
2061 struct user_arg_ptr argv = { 1846 struct user_arg_ptr argv = {
2062 .is_compat = true, 1847 .is_compat = true,
2063 .ptr.compat = __argv, 1848 .ptr.compat = __argv,
2064 }; 1849 };
2065 struct user_arg_ptr envp = { 1850 struct user_arg_ptr envp = {
2066 .is_compat = true, 1851 .is_compat = true,
2067 .ptr.compat = __envp, 1852 .ptr.compat = __envp,
2068 }; 1853 };
2069 return do_execveat_common(fd, filenam 1854 return do_execveat_common(fd, filename, argv, envp, flags);
2070 } 1855 }
2071 #endif 1856 #endif
2072 1857
2073 void set_binfmt(struct linux_binfmt *new) 1858 void set_binfmt(struct linux_binfmt *new)
2074 { 1859 {
2075 struct mm_struct *mm = current->mm; 1860 struct mm_struct *mm = current->mm;
2076 1861
2077 if (mm->binfmt) 1862 if (mm->binfmt)
2078 module_put(mm->binfmt->module 1863 module_put(mm->binfmt->module);
2079 1864
2080 mm->binfmt = new; 1865 mm->binfmt = new;
2081 if (new) 1866 if (new)
2082 __module_get(new->module); 1867 __module_get(new->module);
2083 } 1868 }
2084 EXPORT_SYMBOL(set_binfmt); 1869 EXPORT_SYMBOL(set_binfmt);
2085 1870
2086 /* 1871 /*
2087 * set_dumpable stores three-value SUID_DUMP_ 1872 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2088 */ 1873 */
2089 void set_dumpable(struct mm_struct *mm, int v 1874 void set_dumpable(struct mm_struct *mm, int value)
2090 { 1875 {
>> 1876 unsigned long old, new;
>> 1877
2091 if (WARN_ON((unsigned)value > SUID_DU 1878 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2092 return; 1879 return;
2093 1880
2094 set_mask_bits(&mm->flags, MMF_DUMPABL !! 1881 do {
>> 1882 old = ACCESS_ONCE(mm->flags);
>> 1883 new = (old & ~MMF_DUMPABLE_MASK) | value;
>> 1884 } while (cmpxchg(&mm->flags, old, new) != old);
2095 } 1885 }
2096 1886
2097 SYSCALL_DEFINE3(execve, 1887 SYSCALL_DEFINE3(execve,
2098 const char __user *, filename 1888 const char __user *, filename,
2099 const char __user *const __us 1889 const char __user *const __user *, argv,
2100 const char __user *const __us 1890 const char __user *const __user *, envp)
2101 { 1891 {
2102 return do_execve(getname(filename), a 1892 return do_execve(getname(filename), argv, envp);
2103 } 1893 }
2104 1894
2105 SYSCALL_DEFINE5(execveat, 1895 SYSCALL_DEFINE5(execveat,
2106 int, fd, const char __user *, 1896 int, fd, const char __user *, filename,
2107 const char __user *const __us 1897 const char __user *const __user *, argv,
2108 const char __user *const __us 1898 const char __user *const __user *, envp,
2109 int, flags) 1899 int, flags)
2110 { 1900 {
>> 1901 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
>> 1902
2111 return do_execveat(fd, 1903 return do_execveat(fd,
2112 getname_uflags(fil !! 1904 getname_flags(filename, lookup_flags, NULL),
2113 argv, envp, flags) 1905 argv, envp, flags);
2114 } 1906 }
2115 1907
2116 #ifdef CONFIG_COMPAT 1908 #ifdef CONFIG_COMPAT
2117 COMPAT_SYSCALL_DEFINE3(execve, const char __u 1909 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2118 const compat_uptr_t __user *, argv, 1910 const compat_uptr_t __user *, argv,
2119 const compat_uptr_t __user *, envp) 1911 const compat_uptr_t __user *, envp)
2120 { 1912 {
2121 return compat_do_execve(getname(filen 1913 return compat_do_execve(getname(filename), argv, envp);
2122 } 1914 }
2123 1915
2124 COMPAT_SYSCALL_DEFINE5(execveat, int, fd, 1916 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2125 const char __user *, f 1917 const char __user *, filename,
2126 const compat_uptr_t __ 1918 const compat_uptr_t __user *, argv,
2127 const compat_uptr_t __ 1919 const compat_uptr_t __user *, envp,
2128 int, flags) 1920 int, flags)
2129 { 1921 {
>> 1922 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
>> 1923
2130 return compat_do_execveat(fd, 1924 return compat_do_execveat(fd,
2131 getname_ufl !! 1925 getname_flags(filename, lookup_flags, NULL),
2132 argv, envp, 1926 argv, envp, flags);
2133 } 1927 }
2134 #endif <<
2135 <<
2136 #ifdef CONFIG_SYSCTL <<
2137 <<
2138 static int proc_dointvec_minmax_coredump(cons <<
2139 void *buffer, size_t *lenp, l <<
2140 { <<
2141 int error = proc_dointvec_minmax(tabl <<
2142 <<
2143 if (!error) <<
2144 validate_coredump_safety(); <<
2145 return error; <<
2146 } <<
2147 <<
2148 static struct ctl_table fs_exec_sysctls[] = { <<
2149 { <<
2150 .procname = "suid_dumpa <<
2151 .data = &suid_dumpa <<
2152 .maxlen = sizeof(int) <<
2153 .mode = 0644, <<
2154 .proc_handler = proc_dointv <<
2155 .extra1 = SYSCTL_ZERO <<
2156 .extra2 = SYSCTL_TWO, <<
2157 }, <<
2158 }; <<
2159 <<
2160 static int __init init_fs_exec_sysctls(void) <<
2161 { <<
2162 register_sysctl_init("fs", fs_exec_sy <<
2163 return 0; <<
2164 } <<
2165 <<
2166 fs_initcall(init_fs_exec_sysctls); <<
2167 #endif /* CONFIG_SYSCTL */ <<
2168 <<
2169 #ifdef CONFIG_EXEC_KUNIT_TEST <<
2170 #include "tests/exec_kunit.c" <<
2171 #endif 1928 #endif
2172 1929
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.